Best AI Models in 2026
Rankings, comparisons, and deep dives, updated monthly as new models ship.
- Smartest overallClaude Fable 5.2
- CodingClaude Opus 5.5
- Hardest reasoningGPT-6 Astra
- ImagesChatGPT Images 2.6
- Best free pickClaude Sonnet 5.5
Stop tab-hopping between AI apps
Chat, search, PDFs and images in one app for Mac, iPhone and iPad.
Free to start·4.7★ from 27,000+ reviews
Rankings show who wins on average. Your prompt isn't average. Ask ChatGPT, Claude and Gemini in one window and keep the best answer.
Download Fello AIAnthropic released Claude Fable 5.2 on November 9, and it now leads Artificial Analysis's Intelligence Index v5 at 64.6, the highest score published, ahead of Google's Gemini 4 Carbon at 60.0 and Claude Opus 5.5 at 59.1. It takes the accuracy crown and podium places in chat, coding and problem solving, but it costs twice Opus 5.5's token price, $26.54 an index task at max effort, and it is not on Claude's free plan. Our guide to AI benchmarks explains how that index is built.
Most crowns did not move. Claude Opus 5.5 keeps coding on Arena's WebDev board, GPT-6 Astra keeps problem solving on ARC-AGI-2, and Claude Fable 5.1, now a Legacy model, keeps writing because no prose board has rated 5.2 yet. Video changes hands: Google's Gemini Omni 1.2 Flash, released November 5, entered first on both Arena video boards and takes the crown from MiniMax H3. Claude Mythos 5.2, the same model without Fable 5.2's cyber and biology safeguards, goes only to verified organizations, so it takes no crown. Gemini 4 Carbon, open only to cyber defenders until Google made it generally available on November 10, now counts, but it leads none of the boards our crowns are decided on.
Among open models, Kimi K3.1 Pro still leads at 46.5 on the index, 13.9 points clear, and DeepSeek V4.1-Flash-1017 stays the open value pick. On images, ChatGPT Images 2.6 now leads both of Artificial Analysis's boards as well. Rankings are updated within 48 hours of any major launch.
Nov 9 Anthropic Claude Fable 5.2 tops Intelligence Index v5 at 64.6 and costs 20% less than 5.1; Mythos 5.2 goes to verified organizations only New
Nov 9 Meta releases Muse Ember, which matches Claude Haiku 5.5 on the index for a third of the cost per task New
Nov 5 Google Google ships Gemini Omni 1.2 Flash, its cheaper video model, in the Gemini app, Flow, Vids and the API New
Nov 4 OpenAI OpenAI releases ChatGPT Images 2.6 with animated GIFs and saved reference images New
Nov 2 Moonshot Moonshot publishes the weights of Kimi K3.1 Pro, now the highest-scoring open model by 13.9 points New
Oct 29 Mistral Mistral publishes the weights of Mistral Large 4 and halves its price, but it ranks only seventh among open models New
Oct 28 Google Gemini 4 Carbon tops Intelligence Index v5 at 59.4, but only cyber defenders can use it New
Oct 27 Artificial Analysis Artificial Analysis launches Intelligence Index v5, which re-scales every score on this page New
Oct 26 OpenAI GPT-6.1 Terra scores 51.0 at half GPT-6.1 Sol's price, in ChatGPT Work and Codex New
Oct 23 Moonshot Kimi K3.1 Pro lands fourth on the Intelligence Index and first on Terminal-Bench 4.0, at $2 / $10 New
Oct 17 DeepSeek DeepSeek V4.1-Flash-1017 scores 45.1 at unchanged prices and takes the open price-performance pick New
Oct 15 Meta publishes the weights of Muse Spark Open, an 816B model under its own licence New
Oct 15 fal fal's FLUX 3 Max, a faster post-trained FLUX 3 Video, enters Artificial Analysis's video boards fourth and second New
Pending Google Gemini 3.5 Pro, still unreleased, and Gemini 4 Argon has overtaken it Delayed
Best AI for Writing
The best AI for writing is Claude Fable 5.1, and the case for it is consistency rather than a single first place.
It is inside the top ten of all three prose boards: third on EQ-Bench Creative Writing v3 at 2155.3, behind GPT-6 Astra and Moonshot's new Kimi K3.1 Pro, third on LiveBench 1.0 Language at 89.5 and seventh on Arena's creative-writing board, the best combined placing of any model. Gemini 4 Argon is second on Arena's text board at 1522, behind only Gemini 4 Carbon, and first on its creative-writing board at 1525, and Google AI Pro subscribers can use it since October 13, but EQ-Bench puts it only 32nd, at 1757.8, and the archived LiveBench 1.0 Language board never rated it.
Claude Opus 5.5 is second on Arena's creative-writing board at 1518, behind Gemini 4 Argon, but it is eighth on EQ-Bench and tenth on LiveBench 1.0 Language; GPT-6 Astra leads EQ-Bench at 2176.9 but sits 39th on Arena's creative-writing board; Kimi K3.1 Pro is second on EQ-Bench at 2172.6, but LiveBench 1.0 Language never rated it; Claude Fable 5 tops LiveBench 1.0 Language at 90.7 and is fourth on Arena's creative-writing board, but 13th on EQ-Bench. If your writing is a work deliverable rather than prose, the GDPval-AA v2.1 professional-deliverables board is a different question, and Claude Opus 5.5 leads it at 1866, ahead of Moonshot's Kimi K3.1 Pro at 1861, Meta's Muse Wonder Full at 1857, Claude Sonnet 5.5 at 1838 and Claude Fable 5.2 at 1831.
Claude Sonnet 5.5 is the free-tier value pick, since anyone can use it on claude.ai, though it scores 83.4 on LiveBench 1.0 Language, sits only 25th on Arena's creative-writing board and has no EQ-Bench rating yet. Claude Fable 5 is the pick if you weight Arena's human votes above the harness scores; it costs $10 / $50 per 1M tokens and is permanently included in Claude Max, Ultra and Team Premium at roughly 50% of regular usage limits. Translation is a separate question with a separate winner, which we work through in our guide to the best AI for translation.
| Model | Best For | Strength | Weakness | Price (per 1M tokens) |
|---|---|---|---|---|
| Claude Fable 5.1 | Best writing overall | #3 EQ-Bench Creative Writing (2155.3), #3 LiveBench 1.0 Language (89.5) and #7 Arena creative writing, the best combined placing of any model | Leads none of the three outright; Gemini 4 Argon leads Arena's creative-writing board and GPT-6 Astra leads EQ-Bench; a Legacy model since Fable 5.2 (November 9) | $10 / $50 |
| Claude Fable 5 | Strong on the human-vote boards | #1 LiveBench 1.0 Language (90.7), #4 Arena creative writing (1501), #5 Arena text (1504) | #13 EQ-Bench; priciest option here | $10 / $50 |
| Kimi K3.1 Pro | Creative fiction and voice | #2 EQ-Bench Creative Writing (2172.6), 4.3 points behind GPT-6 Astra; #2 GDPval-AA v2.1 (1861) | Never rated on LiveBench 1.0 Language; open weights since November 2 | $2 / $10 |
| Claude Sonnet 5.5 | Free everyday writing | Free on claude.ai, 1M context; #4 GDPval-AA v2.1 (1838), behind Opus 5.5, Kimi K3.1 Pro and Muse Wonder Full | 83.4 LiveBench 1.0 Language; #25 Arena creative writing; not yet rated on EQ-Bench | $2 / $10 |
| Claude Opus 5.5 | Second on Arena's creative writing | #2 Arena creative writing (1518) and #4 Arena text (1506); #1 GDPval-AA v2.1 (1866) | #8 EQ-Bench and #10 LiveBench 1.0 Language | $4 / $20 |
| Gemini 4 Argon | Top-voted model on Arena you can use | #2 Arena text (1522) and #1 Arena creative writing (1525) | #32 on EQ-Bench (1757.8); unrated on LiveBench 1.0 Language | $19.99/mo Google AI Pro; $2 / $10 API intro to Nov 30 |
| GPT-5.5 | Fact-anchored business writing | Documented factual-reliability gains over GPT-5.4 | Reasoning tiers now marked deprecated | $5 / $30 |
| Gemini 3.8 Flash | Bulk drafts at scale | Intelligence Index 30.8 (v5), 308.4 tok/s output, 1M context | Agent-tuned rather than prose-tuned; intro price doubles January 1, 2027 | $0.75 / $3.75 |
The writing crown stays with Claude Fable 5.1, now a Legacy model, because Claude Fable 5.2, released November 9, has not yet been rated on EQ-Bench, LiveBench 1.0 Language or Arena's creative-writing board; we will re-rank as soon as it is. Since the last update Kimi K3.1 Pro edged up to 1861 on GDPval-AA v2.1, still second behind Opus 5.5, and Fable 5.2 entered that board at 1831. On Arena's creative-writing board, re-read for this update, Gemini 4 Argon has taken first place at 1525 from Claude Opus 5.5, now second at 1518, and Fable 5.1 moved up to seventh. Earlier this month Arena re-ranked its creative-writing board, with Fable 5.1 up to eighth and Gemini 4 Carbon, then open only to cyber defenders, entering third. In October, Kimi K3.1 Pro replaced Kimi K3 on the podium after EQ-Bench rated it second, Claude Opus 5.5 took first place on Arena's creative-writing board and EQ-Bench rated Gemini 4 Argon 32nd.
Best AI for Chat & Daily Assistant
The best AI for everyday chat is GPT-6, and the honest reason is reach rather than board position. Since October 7 it is what ChatGPT answers with by default: GPT-6 Sol on Plus, Pro, Business and Enterprise, and GPT-6 Luna on Free and Go, rolling out from October 8. That makes it the assistant most people can actually open.
On raw human preference it is far from the lead. GPT-6 Sol sits #67 on Arena's text leaderboard at 1457, below its predecessor GPT-5.6 Sol at #22, while Google's Gemini 4 Carbon leads at 1526, Gemini 4 Argon is second at 1522, the highest on Google AI Pro, Meta's Muse Wonder is third at 1511 and Claude Opus 5.5 fourth at 1506; Arena has not rated Claude Fable 5.2 on text yet. Artificial Analysis has not scored GPT-6 Sol on its new Intelligence Index v5, where GPT-6 Luna reads 27.9; on v4.3.2 Sol read 47.6. In the API, Sol costs $2 / $10 and Luna $0.10 / $0.50 per 1M tokens.
| Model | Best For | Strength | Weakness | Price |
|---|---|---|---|---|
| GPT-6 | Everyday chat, ChatGPT's default | The assistant most people can open: Sol on paid plans, Luna on Free and Go | GPT-6 Sol only #67 on Arena text | Free / $20/mo Plus; API $0.10 / $0.50 (Luna) to $2 / $10 (Sol) |
| Gemini 4 Argon | Top-voted conversation | #2 on Arena text (1522); 4% hallucination rate on AA-Omniscience, the lowest among leading models | Arena rating still preliminary; Google AI Pro or above in the Gemini app | $19.99/mo Google AI Pro; $2 / $10 API intro to Nov 30 |
| Claude Fable 5.2 | Highest-scoring conversation you can use | 64.6 on Intelligence Index v5, the highest score published, and 79% factual accuracy on AA-Omniscience | Not on the Free plan; Pro pays with usage credits after the launch offer ends November 16; not yet rated on Arena text; $26.54 an index task at max effort | $20/mo Pro (credits); $8 / $40 API |
| Claude Opus 5.5 | Anthropic's everyday flagship | 59.1 on Intelligence Index v5 and #4 on Arena text (1506); the model Anthropic tells most users to start with | Not on the Free plan; usage-limited on Pro | $20/mo Pro, $4 / $20 API |
| Claude Sonnet 5.5 | Free Claude chat | Free on claude.ai; 54.5 on Intelligence Index v5 at max effort, sixth among distinct models | Only #46 on Arena text; the apps run it at Medium effort, where it scores 30.8 | Free / $20/mo Pro, $2 / $10 API |
| GPT-5.5 Instant | Hallucination-sensitive daily work | 52.5% fewer hallucinated claims vs 5.3 Instant | Reasoning tiers now marked deprecated | $20/mo Plus; API $5 / $30 |
| Gemini 3.6 Flash | Fast, cheap, multimodal | 1M context, #25 on Arena text | Needs Google AI Plus in the Gemini app since October 9; 3.8 Flash outscores it at the same price | AI Plus / $0.75 / $3.75 API |
| Fello AI | Many leading models, one app | ChatGPT + Claude + Gemini + Grok + DeepSeek and more on Mac, iPhone and iPad | Routed via app, not direct | From $11.99/mo in the US (prices vary by country) |
GPT-6 keeps the chat pick, because it is what ChatGPT answers with by default: GPT-6 Sol on paid plans and GPT-6 Luna on Free and Go. Since the last update Anthropic released Claude Fable 5.2 on November 9, which scores 64.6 on Intelligence Index v5 and takes third place from Claude Opus 5.5 as the highest-scoring model you can use, though Arena has not rated it on text yet. On Arena's text board Gemini 4 Argon is still the top model you can use, at 1522, and Meta's Muse Wonder moved ahead of Opus 5.5 into third. In October, the pick moved from GPT-5.6 to GPT-6 on October 7, Argon took second place when it reached Google AI Pro and Ultra on October 13, and on October 29 OpenAI tuned GPT-6 in Chat for Free and Go users to give shorter answers.
Best AI for Images
The best AI for image generation is ChatGPT Images 2.6, which OpenAI released for every ChatGPT plan, free included, on November 4, adding animated GIFs and saved reference images, and OpenAI's models lead all four image boards we track.
GPT-Image-2.5 Sunburst, the precision model from the previous release and still in OpenAI's API, leads both Arena boards, text-to-image at 1429 and image editing at 1520. Artificial Analysis has now rated 2.6, and GPT-Image-2.6 Sunburst leads both of its boards, AA-Image-T2I v2.1 at 1259 and AA-Image-Editing v2.1 at 1208, so OpenAI holds first place on all four. On Arena, 2.6 Sunburst (Medium) is second on text-to-image at 1406 and third on editing at 1477, and OpenAI holds Arena's top five text-to-image spots. OpenAI calls 2.6's quality similar to 2.5 and sells it on speed, GIFs and reference images. MAI-Image-2.6 is the runner-up: sixth on Arena text-to-image at 1329, eighth on Arena editing at 1428 and seventh on Artificial Analysis's text-to-image board at 1168, behind Microsoft's own editing-only MAI-Image-2.7-Preview, which is fourth on both editing boards.
Third is Google's Nano Banana 2.1, released on October 6 and the highest-ranked Google model: sixth on both Artificial Analysis boards, at 1176 on text-to-image and 1151 on editing, ahead of MAI-Image-2.6 on both, and a point behind it on both Arena boards, at 1328 and 1427. SpaceXAI's Grok Imagine Image 2.0, released in August and available in Grok Imagine and the xAI API, is fourth: it still beats Nano Banana 2.1 on Arena's editing board, 1437 to 1427, but scores 1319 on Arena text-to-image and is eighth on Artificial Analysis's text-to-image board, and it stays the pick for the fewest restrictions.
The best image model you can download is Alibaba's Qwen-Image-2.1, released on September 20: the highest-scoring open-weight model on both Arena boards and, at 23rd, on Artificial Analysis's text-to-image board, where it is now just ahead of Black Forest Labs' FLUX 3 Image Klein. On Artificial Analysis's editing board it is 24th, one place behind Bria's non-commercial Fibo 2.0. Its licence allows research and evaluation only, so it is not a free commercial option. Our full breakdown of the new release is in ChatGPT Images 2.6.
| Model | Best For | Strength | Weakness | Price |
|---|---|---|---|---|
| ChatGPT Images 2.6 | Images with readable text, GIFs | OpenAI leads all four boards: GPT-Image-2.6 Sunburst is #1 on Artificial Analysis text-to-image (1259) and editing (1208); GPT-Image-2.5 Sunburst, still in the API, is #1 on Arena text-to-image (1429) and editing (1520), with 2.6 Sunburst #2 (1406) and #3 (1477) there | OpenAI calls its quality similar to 2.5, and low- and medium-quality API images use 3-4x the tokens | Every ChatGPT plan, including free |
| MAI-Image-2.6 | Editing an image you already have | #6 on Arena text-to-image (1329) and #8 on Arena editing (1428); #7 on Artificial Analysis text-to-image (1168) and #8 on its editing board (1149) | Public preview, not in Copilot or Bing yet; its successor MAI-Image-2.7-Preview, editing only so far, is fourth on both editing boards | MAI Playground / Microsoft Foundry |
| Nano Banana 2.1 | Google's everyday image model | #6 on both Artificial Analysis boards (1176 text-to-image, 1151 editing), ahead of MAI-Image-2.6, and 1328 on Arena text-to-image, the highest of any Google model; half Nano Banana 2's API price per image | 1427 on Arena image editing, behind Grok Imagine Image 2.0; input tokens cost three times as much as on Nano Banana 2 | Gemini app / AI Studio; $0.0336 per 1K image |
| Grok Imagine Image 2.0 | Fewest restrictions, Spicy Mode | 1437 on Arena image editing, ahead of Nano Banana 2.1, and #8 on Artificial Analysis text-to-image (1166) | 1319 on Arena text-to-image and outside the top 18 on Artificial Analysis image editing | $30/mo Premium (X Subscription); xAI API |
| Seedream 5.0 Pro | Multilingual text + region editing | 10+ languages incl. Arabic RTL, lasso and layer editing | Around tenth on the independent boards; copyright cloud | BytePlus / Magnific |
| Midjourney v8 | Stylized art, illustration | Aesthetic baseline most artists prefer | Weaker on text in image | $10-$120/mo |
| Qwen-Image-2.1 | Best image model you can download | Highest-scoring open-weight model on both Arena boards (1363 editing, 1227 text-to-image) and on Artificial Analysis text-to-image (#23); #24 on Artificial Analysis editing, one place behind Fibo 2.0; 7B generator, transparent PNGs, up to 10 reference images | Qwen Research License, research or evaluation only, no commercial use | Free download (Hugging Face) |
| Qwen-Image-3.0-Pro | Dense, text-heavy layouts | 1254 on Arena text-to-image and #19 Artificial Analysis text-to-image (1096); prompts up to 4.5k tokens, text in 12 languages | Hosted only, no weights; not rated on Arena image editing | Qwen Chat; $0.04-$0.075 per image in the API |
| Muse Image | Image editing, Meta ecosystem | #12 on Artificial Analysis image editing (1127) and #10 on its text-to-image board | Off the podium; 1266 on Arena text-to-image and 1400 on Arena editing | Meta AI app |
ChatGPT Images 2.6 keeps the image pick, and since the last update Artificial Analysis has rated it: GPT-Image-2.6 Sunburst entered first on both its text-to-image and editing boards, ahead of OpenAI's own 2.5 Sunburst, which still leads both Arena boards. Nano Banana 2.1 now sits ahead of MAI-Image-2.6 on both Artificial Analysis boards and a point behind it on both Arena boards, so the podium order stands. ChatGPT Images 2.6 took over the pick from 2.5 on November 4, and earlier this month the third podium place moved from Grok Imagine Image 2.0 to Nano Banana 2.1.
In October, Artificial Analysis moved both its image boards to v2.1, Google released Nano Banana 2.1 on October 6, Microsoft's editing-only MAI-Image-2.7-Preview entered near the top of both editing boards, third on Arena's at 1471 and second on Artificial Analysis's at 1193, and Grok Imagine Image 2.0 took third place from Muse Image, the place it has now lost to Nano Banana 2.1.
Best AI for Video
The best AI for video generation is Gemini Omni 1.2 Flash, which Google released on November 5, and it went straight to first on both of Arena's video boards, though on preliminary ratings and before Artificial Analysis has rated it.
It leads Arena's text-to-video board at 1540, ahead of its predecessor Omni 1.1 Flash at 1513, and its image-to-video board at 1503, ahead of ByteDance's Dreamina Seedance 2.5 at 1494 and MiniMax H3 at 1493, and it is third on Arena's new video-editing board at 1396. It makes clips of 3 to 15 seconds with native audio, extendable to 30, from up to six reference images, for about $0.07 a second at 720p against roughly $0.10 for 1.1, and it runs in the Gemini app on any Google AI plan, in Flow and in Google Vids. From December 3, API requests to Omni 1.1 Flash are routed to 1.2. Our Gemini Omni Flash guide has the detail.
MiniMax H3, our pick through October, is the runner-up and the open-weight choice: third on Arena's image-to-video board, third on Artificial Analysis's text-to-video board at 1161 and fourth on its image-to-video board at 1133, but only 11th on Arena's text-to-video board at 1456. It makes 2K clips of 4 to 15 seconds with native stereo audio, from $0.13 a second at 2K, and the weights are on Hugging Face under MiniMax's own community licence, without the 2K upscaler and with an application required in the US, EU, UK and South Korea.
On Artificial Analysis, Alibaba's Wan 3.0 leads text-to-video at 1175 and Dreamina Seedance 2.5 leads image-to-video at 1161, with Utopai X, which has no public API, second on both; Wan 3.0 also leads Arena's video-editing board at 1424. If you need longer takes inside Google's tooling, Veo 3.1 still runs in the Gemini app, AI Studio and Vertex AI with native audio and 1080p output.
| Model | Best For | Strength | Weakness | Price |
|---|---|---|---|---|
| Gemini Omni 1.2 Flash | Best AI video overall | #1 Arena text-to-video (1540) and image-to-video (1503), #3 Arena video editing (1396); 3-15s clips with native audio, extension to 30s, up to six reference images | Both Arena ratings preliminary; not yet rated by Artificial Analysis; 15-second clips in the Gemini app need Ultra | ~$0.07/sec at 720p; Gemini app (any Google AI plan), Flow, Vids, Gemini API |
| MiniMax H3 | Open-weight video | #3 Arena image-to-video (1493), #3 and #4 on Artificial Analysis (1161 text-to-video, 1133 image-to-video); 2K clips of 4-15s with native stereo audio | #11 on Arena text-to-video (1456); the open weights exclude the 2K upscaler and need an application in the US, EU, UK and South Korea | $0.13/sec at 2K |
| Gemini Omni 1.1 Flash | Being retired | #2 Arena text-to-video (1513), #4 Arena image-to-video (1487) | Replaced by Omni 1.2 Flash on November 5, with API requests routed to 1.2 from December 3; only #12 on Artificial Analysis text-to-video (1128) and #17 on image-to-video (1044) | ~$0.10/sec at 720p; Gemini API until December 3 |
| Wan 3.0 | Video from documents and slides | #1 Artificial Analysis text-to-video (1175), #1 Arena video editing (1424), #5 Arena image-to-video (1482); 2-30s, up to 1080p, Omni-Reference input | API-only, no published weights; #8 on Arena text-to-video | $0.05-$0.20/sec |
| Dreamina Seedance 2.5 | ByteDance challenger | #1 Artificial Analysis image-to-video (1161), #4 Artificial Analysis text-to-video (1155), #2 Arena image-to-video (1494) | #7 Arena text-to-video (1479); ByteDance ecosystem, limited Western access | Dreamina / BytePlus |
| Muse Video | Meta ecosystem video | #10 on Arena text-to-video at 1457 | Newest of the group, thin tooling | Meta AI app |
| Veo 3.1 | Longer production clips | Native audio, 1080p, strong physics consistency | #14 on Arena text-to-video, not the quality leader | Google AI Pro / Ultra |
| Kling 3.0 / 3.0 Turbo | Fast iteration at lower cost | Native 4K, 60fps, 15-second clips; Turbo shipped June 17 | Outside the top 16 on Arena text-to-video | From $10/mo |
| Luma Ray 3 | Photoreal scenes | Strong realism for landscapes | Smaller community | Free / from $9.99/mo |
Gemini Omni 1.2 Flash takes the video crown from MiniMax H3. Since the last update Arena refreshed both its video boards for the first time since September 21, and Omni 1.2 Flash entered first on both, at 1540 on text-to-video and 1503 on image-to-video, on preliminary ratings; H3 fell to third on image-to-video and 11th on text-to-video. Artificial Analysis has not rated Omni 1.2 Flash yet, and if its boards disagree we will revisit the call. Arena also added a video-editing board, which Wan 3.0 leads.
Earlier this month Google released Omni 1.2 Flash on November 5, at about $0.07 a second against roughly $0.10 for 1.1, and it now runs in the Gemini app, Flow and Vids. In October the crown moved from Omni 1.1 Flash to H3 once Artificial Analysis first rated Omni 1.1 Flash and added audio to its video boards.
Best AI for Coding
The best AI for coding is Claude Opus 5.5, which Anthropic shipped on September 22. It still leads Arena's human-voted WebDev board, but its newer sibling, Claude Fable 5.2, now leads FrontierCode 2.0, and Google's Gemini 4 Carbon leads LiveBench 2.0's coding suite.
Claude Fable 5.2, released November 9, scores 49.1% on FrontierCode 2.0, the coding test inside Intelligence Index v5, against Opus 5.5's 46.4% and GPT-6 Astra's 42.4%, and Arena's AutoEval estimate puts it at 1832 on WebDev, above Opus 5.5, though it has no human votes there yet. It costs twice as much per token, $8 / $40, and on Artificial Analysis's Terminal-Bench 4.0 run it reaches 61.6%, behind Moonshot's Kimi K3.1 Pro at 67.2%, Google's Gemini 4 Carbon at 65.2% and Claude Sonnet 5.5 at 63.6%. Opus 5.5 keeps the crown on human votes: it leads Arena's WebDev board at 1818, ahead of Sonnet 5.5 at 1774 and GPT-6 Astra at 1767. On LiveBench 2.0, live since November 11, it scores 72.8 on the agentic suite against Sonnet 5.5's 63.4, and 87.1 on Coding, where Gemini 4 Carbon leads at 91.0. On price the Claude models are closer than their token rates suggest.
Sonnet 5.5 is $2 / $10 per 1M tokens against Opus 5.5's $4 / $20, but on Intelligence Index v5 Artificial Analysis puts a max-effort task at $10.02 against Opus 5.5's $10.04, and Opus 5.5 at xhigh, 57.4 for $6.05 a task, scores higher than Sonnet 5.5 at max, 54.5, for 60% of the cost. Fable 5.2 at high effort scores 61.8 for $9.08 a task, above Opus 5.5 at max for slightly less. Opus 5.5 also leads Arena's image-to-WebDev board at 1750, ahead of Sonnet 5.5 at 1743, Claude Fable 5.1 at 1724 and GPT-6 Astra at 1722; Astra matches Opus 5.5's 59.6% on Terminal-Bench 4.0 at xhigh, at $10 / $50.
GPT-6.1 Sol is the cheap OpenAI option: sixth on WebDev, 56.1% on Terminal-Bench 4.0 and 54.6 on Intelligence Index v5 for $1.27 an index task at max effort, and ninth on LiveBench 2.0 Coding at 82.1. GPT-6.1 Terra, released October 26, halves that token price to $1 / $5 and scores 51.0 for $0.96 a task, with 55.6% on Terminal-Bench 4.0. Gemini 4 Argon leads Google's own DeepSWE v1.1 table at 77.9%, but Artificial Analysis measures 57.1% on Terminal-Bench 4.0, Arena puts it eleventh on WebDev, and since October 13 it is on the paid API and in Antigravity for Google AI Ultra subscribers.
Claude Haiku 5.5, released October 7 at $0.10 / $0.50, is the cheap Claude for high-volume sub-agent work: 33% on Artificial Analysis's Terminal-Bench 4.0 run, level with GLM 5.3 Flash, and 77.6 on LiveBench 2.0 Coding, just above Sonnet 5.5's 77.2, though Arena puts it only 36th on WebDev. Among open weights, Kimi K3.1 Pro, whose weights Moonshot published on November 2, now leads on Intelligence Index v5 at 46.5, on Arena WebDev in tenth and on Terminal-Bench 4.0, where its 67.2% is first overall. DeepSeek V4.1-Flash-1017, released October 17, is the open value pick at 32.6 and 16th on WebDev, ahead of Kimi K3 in 17th; GLM-5.3 reaches 41.9% on Terminal-Bench 4.0, and Xiaomi's MiMo-V2.6-Pro takes 34.8% there under plain MIT at $0.26 an index task.
Artificial Analysis has retired both its Coding Index and its Coding Agent Index, so the two composite coding rankings this page used to quote no longer exist.
| Model | Best For | Strength | Weakness | Price (per 1M tokens) |
|---|---|---|---|---|
| Claude Opus 5.5 | Best coding overall | #1 Arena WebDev (1818) and #1 image-to-WebDev (1750); #2 FrontierCode 2.0 (46.4%) and 59.1 on Intelligence Index v5, both behind Claude Fable 5.2; 72.8 on LiveBench 2.0's agentic suite, nine points above Sonnet 5.5 | Behind Sonnet 5.5 on Terminal-Bench 4.0 (59.6% to 63.6%); fifth on LiveBench 2.0 Coding (87.1), behind Gemini 4 Carbon's 91.0 | $4 / $20 |
| Claude Sonnet 5.5 | Value pick for coding | #3 Terminal-Bench 4.0 (63.6%) and #2 Arena WebDev (1774), at half Opus 5.5's token price | 14th on LiveBench 2.0 Coding (77.2) and nine points behind Opus 5.5 on its agentic suite; on Intelligence Index v5 Opus 5.5 at xhigh outscores it at max for less per task | $2 / $10 |
| Claude Fable 5.2 | Hardest agentic coding | #1 FrontierCode 2.0 (49.1%); 64.6 on Intelligence Index v5, the highest score published, or 61.8 for $9.08 an index task at high effort; 1832 on Arena WebDev by AutoEval | Twice Opus 5.5's token price; no human votes on Arena yet; 61.6% Terminal-Bench 4.0, fourth among distinct models; Anthropic still tells most users to start on Opus 5.5 | $8 / $40 |
| GPT-6 Astra | OpenAI's strongest coder | #4 Arena WebDev (1767) and #4 image-to-WebDev (1722); #3 FrontierCode 2.0 (42.4%); 59.6% Terminal-Bench 4.0 at xhigh | 2.5x Opus 5.5's price; behind Sonnet 5.5 on Terminal-Bench and behind Gemini 4 Carbon on LiveBench 2.0 Coding (87.5 to 91.0) | $10 / $50 |
| Kimi K3.1 Pro | Strongest terminal coder | #1 Terminal-Bench 4.0 (67.2%); 46.5 on Intelligence Index v5 at $3.68 an index task | #10 Arena WebDev (1688); open weights since November 2, but at 2.8T parameters a data-centre model | $2 / $10 |
| Claude Fable 5.1 | Legacy since November 9 | #5 Arena WebDev (1745) and #3 image-to-WebDev (1724) | Replaced by Fable 5.2, which costs less per token; Artificial Analysis no longer publishes its scores | $10 / $50 |
| Claude Opus 5 | Superseded by Opus 5.5 | 49.0% Terminal-Bench 4.0 and #9 Arena WebDev and #6 image-to-WebDev | Costs more per token than Opus 5.5 and scored seven points lower on v4.3.2; not scored on Intelligence Index v5 | $5 / $25 |
| Claude Fable 5 | Hardest long-horizon agentic work | Intelligence Index 49.6 (v4.3.2); 42.4% Terminal-Bench 4.0; 1M context | The priciest per index task on v4.3.2 at $8.75; #10 on Arena image-to-WebDev | $10 / $50 |
| Qwen3.8-Max-0902 | Brief holder of Arena's WebDev board | #13 Code Arena: WebDev (1677); 33.6 on Intelligence Index v5; 2.4T parameters, 1M context | QwenCloud API only with no consumer front-end; lost the WebDev lead within days | $2 / $6 |
| DeepSeek V4.1-Flash-1017 | Open value pick | 32.6 on Intelligence Index v5, second among open models after Kimi K3.1 Pro, at $0.60 a task; #16 Arena WebDev (1664) and #20 Agent Arena; MIT | 33.8% on Terminal-Bench 4.0; 552B/16B active to self-host | $0.15 / $0.60 off-peak |
| GPT-5.6 Sol | OpenAI's cheaper flagship | Intelligence Index 47.0 (v4.3.2); 39.9% Terminal-Bench 4.0 | Superseded by GPT-6 Sol, which scores higher at half the price; eval-gaming flagged by METR | $4 / $20 |
| GPT-6 Sol | Cheap GPT-6 coding in Codex | Intelligence Index 47.6 (v4.3.2) and 43.9% Terminal-Bench 4.0, both above GPT-5.6 Sol, at half its price | #14 Arena WebDev (1674) | $2 / $10 |
| GPT-6.1 Sol | Cheapest top-tier model per task | 54.6 on Intelligence Index v5, #6 Arena WebDev (1743) and 56.1% Terminal-Bench 4.0 at $1.27 an index task, or 50.3 for $0.59 at high effort; cached input at $0.10 | ninth on LiveBench 2.0 Coding (82.1); in Codex and ChatGPT Work only, not in Chat | $2 / $10 |
| GPT-6.1 Terra | OpenAI's half-price coder | 51.3 on Intelligence Index v5 at $0.96 an index task; 55.6% Terminal-Bench 4.0; fourth among distinct models on FrontierCode 2.0 (42.3%) | In ChatGPT Work and Codex only, not in Chat, and 56.2 on LiveBench 2.0's agentic suite | $1 / $5 |
| Gemini 4 Argon | Long software tasks | 77.9% DeepSWE v1.1 on Google's own figures; 50.4 on Intelligence Index v5 | 57.1% Terminal-Bench 4.0 and #11 Arena WebDev (1683) | $2 / $10 intro to Nov 30, $4 / $20 standard |
| Gemini 4 Carbon | Long coding and cyber defense | 60.0 on Intelligence Index v5, second only to Claude Fable 5.2; #2 Terminal-Bench 4.0 (65.2%) | In the Gemini app on Google AI Ultra only; seventh among distinct models on FrontierCode 2.0 (40.6%) | $2 / $10 until Nov 30, then $4 / $20 |
| Muse Spark 1.3 | Cheap agentic coding | Intelligence Index 48.1 (v4.3.2) at $1.25 / $4.25, and $1.60 per index task | 33.3% on Terminal-Bench 4.0; the coding wins come from Meta's own chart, measured on a partners-only max variant; not yet scored on v5 | $1.25 / $4.25 |
| Muse Wonder Full | Meta's strongest model, inside Muse Code | 50.1 on Intelligence Index v5, just below Gemini 4 Argon; #3 GDPval-AA v2.1 (1857); 57.1% Terminal-Bench 4.0 | Muse Code subscription only ($50 and $200 plans), no API; seventh on Arena WebDev (1724) | Muse Code, from $50/mo |
| Muse Wonder | Meta's frontier model on the API | 43.6 on Intelligence Index v5; 52.0% Terminal-Bench 4.0; 1M context | Same $4 / $20 as Opus 5.5, which scores 15.5 points higher; $6.95 an index task | $4 / $20 |
| Grok 4.7 | Cheap value coder | 46.3% CursorBench 4.0 and 71.0% DeepSWE v1.1 on SpaceXAI's own figures; 35.2 on Intelligence Index v5 | 24.7% on Terminal-Bench 4.0; $6.18 an index task; prompts of 200K tokens and up re-bill the whole request at double | $2 / $6 |
| Gemini 3.8 Flash | Agent coding at scale | 30.8 on Intelligence Index v5; 308.4 output tokens per second | 19.7% on Terminal-Bench 4.0; intro price doubles January 1, 2027 | $0.75 / $3.75 |
| GLM 5.3 Flash | Smallest serious coder you can host | 27.8 on Intelligence Index v5 at $0.58 an index task and 32.8% Terminal-Bench 4.0; MIT, 320B/18B active, against 552B for DeepSeek V4.1-Flash | GLM-5.3 reaches 41.9% on the same harness; no consumer product | Open weights (MIT) |
| Claude Haiku 5.5 | Cheapest Claude for sub-agents | 36.5 on Intelligence Index v5 at $0.41 an index task; 32.8% Terminal-Bench 4.0, level with GLM 5.3 Flash; 77.6 on LiveBench 2.0 Coding, just above Sonnet 5.5 | #36 Arena WebDev (1595); Anthropic says Sonnet 5.5 and Opus 5.5 remain better for complex agentic coding; prompts above 100K tokens cost five times as much | $0.10 / $0.50 |
| Muse Ember | Haiku-level score for less per task | 36.0 on Intelligence Index v5 at $0.14 an index task, a third of Haiku 5.5's; #34 Arena WebDev (1608, preliminary) | Muse Code and the Meta Model API only, no consumer app and no weights; the price triples above 200K tokens | $0.10 / $0.50 |
| MiMo-V2.6-Pro | Cheapest strong open model per task | 29.9 on Intelligence Index v5 and 34.8% Terminal-Bench 4.0 at $0.26 an index task; MIT, 1.02T/42B active | Only #27 on Arena WebDev (1624), behind DeepSeek V4.1-Flash-1017 and Kimi K3 | $0.435 / $0.87 |
Claude Opus 5.5 keeps the coding crown on Arena's human-voted WebDev board, where it leads at 1818, ahead of Claude Sonnet 5.5 at 1774 and GPT-6 Astra at 1767. Since the last update Anthropic released Claude Fable 5.2 on November 9: it takes FrontierCode 2.0 from Opus 5.5, 49.1% to 46.4%, and Arena's AutoEval puts it at 1832 on WebDev, so it takes second place on the podium, while Sonnet 5.5, now second on WebDev, replaces Astra in third as the value pick. If Fable 5.2's WebDev rating holds once human votes come in, it takes the crown. Meta's Muse Ember, released the same day at $0.10 / $0.50, matches Claude Haiku 5.5 on the index for a third of the cost per task. Earlier this month Moonshot published the weights of Kimi K3.1 Pro, now the strongest open model for coding.
In October, Kimi K3.1 Pro took Terminal-Bench 4.0 at 67.2% from Sonnet 5.5, Claude Haiku 5.5 arrived on October 7 at $0.10 / $0.50 as the cheapest Claude for sub-agent work, GPT-6.1 Terra arrived on October 26 at $1 / $5, Artificial Analysis added FrontierCode 2.0 with Intelligence Index v5 on October 27, and DeepSeek V4.1-Flash-1017 became the highest-placed open model on WebDev.
Best AI for Creativity
The best AI for unfiltered, on-trend creative work is Grok 4.7, and we want to be exact about why. This pick is about the product line, not the prose quality.
Grok carries the fewest content restrictions of any frontier model and the only native real-time X integration, which makes it the one model that will engage with edgy, topical or deliberately provocative briefs that the others decline. SpaceXAI shipped Grok 4.7 on September 21 to the Grok API, Cursor and Grok Build, and since October 3 it also powers Expert mode on grok.com, plus Fast mode for paid subscribers. It is not the best writer, though the picture is more mixed than it was.
EQ-Bench has rated Grok 4.7 and puts it ninth at 2012.2, inside its top ten and far above Grok 4.5 in 57th, but Arena's human voters put 4.7 only 86th on creative writing, behind Grok 4.6 in 58th and the older Grok 4.20-beta1 in 26th. If you are picking on output quality alone, Gemini 4 Argon, on Google AI Pro since October 13, now leads Arena's creative-writing board, with Claude Opus 5.5 second, Gemini 4 Carbon third and Claude Fable 5 fourth, while EQ-Bench puts GPT-6 Astra first at 2176.9, Moonshot's Kimi K3.1 Pro second and Claude Fable 5.1 third. Choose Grok for what it will let you make, not for how well it writes.
| Model | Best For | Strength | Weakness | Price |
|---|---|---|---|---|
| Grok 4.7 | Unfiltered, opinionated, on-trend | Fewest content restrictions, native real-time X grounding | #86 on Arena creative writing, though #9 on EQ-Bench (2012.2) | $2 / $6 API; $30/mo Premium app (X Subscription) |
| Claude Fable 5 | Highest-quality creative prose | #4 Arena creative writing, #1 LiveBench 1.0 Language | Cautious guardrails on edgy briefs | $10 / $50 API |
| Kimi K3.1 Pro | Fiction and distinctive voice | #2 EQ-Bench Creative Writing at 2172.6 | Never rated on LiveBench 1.0 Language; open weights since November 2 | $2 / $10 |
| Claude Opus 5 | Long-form structured creativity | Holds long threads and self-edits; 50.8 on Intelligence Index v4.3.2, not scored on v5 | Most cautious of the group | $20/mo Pro; $5 / $25 API |
| Gemini 3.1 Pro | Multimodal creative | Strong text, image and video chain | Needs Google AI Pro in the Gemini app since October 9 | AI Pro / $2.00-$4.00 API in |
| Grok Imagine (Spicy Mode) | NSFW / adult creative | Most permissive image generation | Niche use case | $30/mo Premium, most likely (xAI names no tier) |
Grok keeps this pick on Grok 4.7, and nothing about the product's permissiveness or its live X access changed, which is what the pick rests on. Since the last update no creative-writing board has rated Claude Fable 5.2, released November 9, so the podium stands. Earlier this month Arena re-ranked its creative-writing board: Grok 4.7 slipped from 78th to 87th, Google's Gemini 4 Carbon, then open only to cyber defenders, entered third, and Claude Fable 5 moved down to fourth, where it keeps its podium place here as the quality pick. In October, Kimi K3.1 Pro replaced Kimi K3 on the podium as the fiction pick after EQ-Bench rated it second, Claude Opus 5.5 took the lead on Arena's creative-writing board, and on October 3 Grok 4.7 started powering Expert mode on grok.com.
Best AI for Accuracy & Research
The best AI for accuracy and research is Claude Fable 5.2, released November 9, which answers 79% of AA-Omniscience questions correctly, the highest factual accuracy Artificial Analysis has measured.
That is the exact column this category is decided on, and Fable 5.2 wins it by a distance: Claude Opus 5.5 reads 66% and Google's Gemini 4 Carbon 58%. It also leads the hallucination-penalised AA-Omniscience index at 66.8, ahead of Carbon at 53.5, and Humanity's Last Exam at 74.3%, against Opus 5.5's 61.4%. Its predecessor, Claude Fable 5.1, held this crown at 68% and is now a Legacy model. The catches are price and caution: Fable 5.2 costs $8 / $40 per 1M tokens, twice Opus 5.5's rate, it is not on Claude's free plan, and when it does not know an answer it guesses more often than Opus 5.5 does, a 59% hallucination rate against 50%.
The value pick is GPT-6.1 Sol, which answers 62% correctly with an AA-Omniscience index of 42.0 at $2 / $10 per 1M tokens and $1.27 an index task, and scores 98.5% on ARC Prize's ARC-AGI-1 for $0.06 a task. That beats Gemini 3.1 Pro, our value pick until early October, on every accuracy measure Artificial Analysis has published for it: it read 55% raw accuracy and an index of 31.9.
Gemini 3.1 Pro keeps its case where the answer has to be current, because it grounds answers in Google Search and is in the Gemini app (Google AI Pro and above since October 9), while GPT-6.1 Sol runs in ChatGPT Work, Codex and the API rather than in ChatGPT's chat window. Gemini 4 Argon is the model to watch: Artificial Analysis measures a 4% hallucination rate, the lowest among leading models, because it declines rather than guesses, but it answers only 48% correctly; since October 13 it is in the Gemini app on Google AI Pro and above.
On grounded search specifically, Arena's search leaderboard is led by Qwen3.8-Max at 1260, level with OpenAI's GPT-5.6 Sol at 1258, with Claude Fable 5.1 sixth and Gemini 3.1 Pro grounding 15th; Arena has not rated Fable 5.2 on search yet. On novel reasoning, GPT-6 Astra leads ARC-AGI-2 and Claude Fable 5.2 leads ARC-AGI-3. Claude Opus 5 holds no accuracy crown because Artificial Analysis measures its hallucination rate at 61%.
| Model | Best For | Key Benchmark | Weakness | Price |
|---|---|---|---|---|
| Claude Fable 5.2 | Highest measured factual accuracy | 79% factual accuracy on AA-Omniscience, the highest Artificial Analysis has measured, 14 points above Claude Opus 5.5; #1 AA-Omniscience index (66.8) and #1 Humanity's Last Exam (74.3%) | 59% hallucination rate, against Gemini 4 Argon's 4%; not on the Free plan; not yet rated on Arena search | $8 / $40 |
| Claude Opus 5.5 | Strong accuracy at half the price | #2 Humanity's Last Exam (61.4%) and #3 AA-Omniscience index (48.1); 66% raw accuracy | 13 points behind Fable 5.2 on raw accuracy | $4 / $20 |
| GPT-6.1 Sol | Best value | 62% raw accuracy and an AA-Omniscience index of 42.0 at $1.27 an index task; 98.5% ARC-AGI-1 at $0.06 a task | 52% hallucination rate; in ChatGPT Work and Codex, not in Chat | $2 / $10 |
| Gemini 4 Argon | Fewest hallucinations | 4% hallucination rate on AA-Omniscience, the lowest among leading models | Only 48% raw accuracy | $19.99/mo Google AI Pro; $2 / $10 API intro |
| Gemini 4 Carbon | On the API and AI Ultra since Nov 10 | #1 AA-Omniscience index (53.5), with an 11% hallucination rate | 58% raw accuracy | $2 / $10 until Nov 30 |
| Gemini 3.1 Pro | Grounded search; Gemini app on AI Pro from Oct 9 | Native Google Search grounding; 98% ARC-AGI-1 at $0.52/task | 55% raw accuracy on AA-Omniscience, below GPT-6.1 Sol; #15 on Arena search | $2.00-$4.00 / $12.00-$18.00 (tiered) |
| Claude Fable 5 | Grounded search | #10 on Arena's search leaderboard, behind GPT-5.6 Sol | No single cheap tier | $10 / $50 (Fable 5) |
| GPT-5.6 Sol | Novel reasoning | #5 ARC-AGI-2 at 92.5%, behind GPT-6 Astra (95%), GPT-6.1 Sol (94.2%), Claude Fable 5.2 (93.8%) and Claude Opus 5.5 (93.3%) | Scheming flagged by METR | $4 / $20 |
| Claude Opus 5 | Hardest unseen problems | 30.2% ARC-AGI-3 on the standard harness, well behind Claude Fable 5.2 (81.2%) and GPT-6 Astra (62.7%) | Artificial Analysis measures a 61% hallucination rate | $5 / $25 |
| Qwen 3.7 Max | Frontier accuracy at value pricing | 92.4 GPQA Diamond, 200 free requests/day | API-only, no chat front-end | $1.25 / $3.75 promo; $2.50 / $7.50 list |
| Claude Opus 4.6 | Honesty under pressure | #1 on Scale SEAL's MASK board at 96.28; Anthropic holds the top 5 | Superseded as a flagship | Legacy Anthropic model |
Claude Fable 5.2 takes the accuracy crown from Claude Fable 5.1, which held it at 68% raw factual accuracy and is now a Legacy model. Since the last update Artificial Analysis rated Fable 5.2 at 79% raw accuracy, 66.8 on the hallucination-penalised AA-Omniscience index, ahead of Google's Gemini 4 Carbon at 53.5, and 74.3% on Humanity's Last Exam, ahead of Claude Opus 5.5's 61.4%. On Arena's search leaderboard, re-read for this update, Qwen3.8-Max has moved level with OpenAI's GPT-5.6 Sol at the top, Fable 5.1 is sixth, and Arena has not rated Fable 5.2 there yet.
In October, GPT-6.1 Sol took the value card from Gemini 3.1 Pro, beating it on every accuracy measure Artificial Analysis publishes, Google announced Gemini 4 Carbon on October 28, and Gemini 4 Argon reached the Gemini app on October 13 with the lowest hallucination rate Artificial Analysis has measured among leading models.
Best AI for Problem Solving
The best AI for hard problem solving is GPT-6 Astra, which took the reasoning boards off GPT-5.6 Sol within days of launching.
It leads ARC-AGI-2 at 95%, the highest anyone has scored against that board's 100% human panel, is second on LiveBench 2.0 Reasoning at 92.7, behind Gemini 4 Argon at 95.1, and sat fourth on the archived LiveBench 1.0 Mathematics board at 96.81, behind Claude Opus 5.5 at 97.08, Claude Fable 5.1 at 97.01 and GPT-6.1 Sol at 96.83. It also reports 97.6% on FrontierMath Tier 4 v2, a number worth carrying with Epoch AI's disclosure that OpenAI funded the benchmark and holds exclusive access to part of it. The catch is price and reach: $10 / $50 per 1M tokens against GPT-6.1 Sol's $2 / $10, and it needs a paid ChatGPT plan.
GPT-6.1 Sol, released September 29, is the value alternative and very nearly a co-leader: 92.6 on LiveBench 2.0 Reasoning, 0.1 behind Astra, and 94.2% on ARC-AGI-2, 0.8 points behind, for $0.25 a task against Astra's $1.12. Astra keeps the crown because it leads ARC-AGI-2 and trails only Gemini 4 Argon on LiveBench Reasoning. Claude Fable 5.2, released November 9, has taken ARC-AGI-3 at 81.2% on the standard harness, against Astra's 62.7% and GPT-6.1 Sol's 52.7%, is third on ARC-AGI-2 at 93.8% and leads Humanity's Last Exam at 74.3% and CritPt, Artificial Analysis's physics-research test, at 35.7%, though LiveBench 2.0 puts it 16th on Reasoning at 88.8. Gemini 4 Argon leads Arena's math category, but on only a few hundred votes, and it leads LiveBench 2.0 Reasoning at 95.1; ARC Prize puts it sixth on ARC-AGI-2 at 91.3% and fourth on ARC-AGI-3 at 49.8%. GPT-5.6 Sol, the old value pick, was fifth on the archived LiveBench 1.0 Reasoning board and fourth on ARC-AGI-2 at 92.5%.
Qwen 3.7 Max is the budget pick for competition-style problems at 97.1 on the February 2026 HMMT index and 44.5 on Apex, with 200 free model requests a day. Claude Opus 5, the old pick for long agentic reasoning chains, is now a legacy model.
| Model | Best For | Key Benchmark | Weakness | Price |
|---|---|---|---|---|
| GPT-6 Astra | Hardest math, science and reasoning | #1 ARC-AGI-2 (95%), #2 LiveBench Reasoning (92.7); #2 ARC-AGI-3 (62.7%) | Needs a paid ChatGPT plan; 5x GPT-6.1 Sol's price | $10 / $50 |
| Claude Fable 5.2 | Novel problems and research-grade science | #1 ARC-AGI-3 (81.2%), #1 Humanity's Last Exam (74.3%), #1 CritPt (35.7%); #3 ARC-AGI-2 (93.8%) | #16 LiveBench Reasoning (88.8); $0.98 an ARC-AGI-2 task against GPT-6.1 Sol's $0.25 | $8 / $40 |
| Claude Opus 5 | Long agentic reasoning chains | 1656 on AA-Briefcase v1.1 and 1730 on GDPval-AA v2.1; 30.2% ARC-AGI-3 | 61% hallucination rate; now a legacy model | $5 / $25 |
| GPT-5.6 Sol | Value STEM flagship | #5 ARC-AGI-2 (92.5%); #5 LiveBench 1.0 Reasoning (91.65) and Mathematics 96.20, both archived | FrontierMath still unpublished; scheming flagged by METR | $100/mo ChatGPT Pro; API $4 / $20 |
| GPT-6.1 Sol | Value pick, near-Astra reasoning | #3 LiveBench Reasoning (92.6), #2 ARC-AGI-2 (94.2%), #3 LiveBench 1.0 Mathematics (96.83); 54.6 on Intelligence Index v5 at $1.27 an index task | 52.7% on ARC-AGI-3 against Astra's 62.7%; in ChatGPT Work and Codex, not in Chat | $2 / $10 |
| Qwen 3.7 Max | Competition math on a budget | 97.1 HMMT 2026 Feb, 44.5 Apex, 200 free requests/day | API-only | $1.25 / $3.75 promo; $2.50 / $7.50 list |
| Claude Fable 5 | Math inside a coding workflow | #3 on Arena's math subcategory (1522), behind Gemini 4 Argon and Claude Opus 5; 96.0 LiveBench 1.0 Mathematics | Priciest option here | $10 / $50 |
| GLM 5.3 Flash | Open-weight problem solving | Intelligence Index 41.8 (v4.3.2) under MIT, almost eight points above GLM-5.2 at less than half the size; 320B/18B active, 1M context | Needs a multi-GPU server, not a workstation; 27.8 on v5; GLM 5.3 scores higher on Z.ai's own figures and is downloadable since August 28, but at more than twice the size and under a bespoke licence | Open weights (MIT) |
GPT-6 Astra keeps the problem-solving crown, by less than a point. It still leads ARC-AGI-2 at 95% against Sol's 94.2%, and on LiveBench 2.0, live since November 11, it is second on Reasoning at 92.7, behind Gemini 4 Argon's 95.1 and 0.1 ahead of GPT-6.1 Sol. Since the last update Anthropic released Claude Fable 5.2 on November 9, and ARC Prize already puts it first on ARC-AGI-3 at 81.2%, ahead of Astra's 62.7%, and third on ARC-AGI-2 at 93.8%, so it replaces Claude Opus 5 in second place; LiveBench 2.0 puts it 16th on Reasoning at 88.8, so the crown stays with Astra. In October, ARC Prize scored GPT-6.1 Sol second on ARC-AGI-2 and Gemini 4 Argon at 91.3%, and Argon took Arena's math category on a few hundred votes.
Best AI Agent
The best AI agent is Gemini Spark if you want the agent in the cloud and Claude Cowork if you want it on your desktop.
These are joint picks rather than a first and a second: Spark runs in a Google Cloud VM, so it keeps working while your laptop is shut, wired into Gmail, Docs and, since early September, Google Photos; Cowork runs on your Mac or Windows machine and drives your local apps and your screen. Spark is no longer the only agent that keeps working in the cloud.
SpaceXAI's Grok Bot (August 11) gives each account a persistent cloud computer that signs into your apps, from a $30/month X Subscription Premium plan; Meta's Muse (September 8) runs errands from a cloud VM of its own and is free up to 100 million tokens a week; and OpenAI dots (September 29) run on GPT-6 Astra with their own cloud computer and browser, from the $100/month Pro plan, which does not include them in the EEA, UK or Switzerland. No independent board rates these products against each other, so the choice is about where the work has to happen rather than which one scores higher.
Price is no longer a tiebreaker either: Spark now reaches the $19.99/month Google AI Pro tier in the US and more than 160 other countries, and Cowork is included at no extra charge on every paid Claude plan from $20/month. ChatGPT Codex Mobile (May 14) is the alternative for coding-agent work. Read the full Gemini Spark vs Claude Cowork comparison.
| Agent | Best For | Where It Runs | Strength | Price |
|---|---|---|---|---|
| Gemini Spark | 24/7 cloud tasks, Workspace workflows | Google Cloud VM (always-on) | Always-on, deep Workspace integration | From $19.99/mo Google AI Pro |
| Claude Cowork | Desktop, app-driving, design + code | Your Mac/Windows desktop | Drives local apps, sees your screen | $20/mo Claude Pro |
| ChatGPT Codex Mobile | Coding agent on phone | OpenAI cloud + iOS/Android | Approve diffs and redirect work from phone | Included in ChatGPT plans |
| Grok Bot | Browser jobs across logged-in apps | One persistent SpaceXAI cloud computer shared by all your Bots | Signs into apps with your own logins and keeps going with your laptop closed | From $30/mo X Subscription Premium (Cursor Pro at $20 for existing subscribers) |
| OpenAI dots | Ongoing responsibilities, not one-off tasks | OpenAI cloud computer and browser | Runs on GPT-6 Astra; keeps working with your computer off; reachable in ChatGPT, Slack and by voice | From $100/mo ChatGPT Pro (not EEA/UK/CH); Business Premium |
| Meta Muse | Free personal errands | A Muse Secure VM in Meta's cloud | Free up to 100M tokens a week; a separate Sentinel agent approves what it sends out | Free; $20/mo Power, $100/mo Maximum |
| Claude Duet | Proactive daily briefings and errands | Its own cloud VM and browser, on iOS, Android and claude.ai | Checks your connected apps every few hours and works with your devices off; asks before it sends, pays, signs or deletes | Research preview, US Claude Pro and Max, waitlist |
Gemini Spark (cloud) and Claude Cowork (desktop) remain joint picks, because this category crowns products, and no new agent product has shipped since the last update. The model layer barely moved: Claude Opus 5.5 still tops Arena's Agent Arena, now at 20.7% confirmed success, with Google's Gemini 4 Carbon second on a preliminary rating, Claude Sonnet 5.5 still leads AA-Briefcase v1.1 at 1816, and on GDPval-AA v2.1 Moonshot's Kimi K3.1 Pro is second at 1861, behind Opus 5.5 at 1866 and four points ahead of Meta's Muse Wonder Full. Arena has not rated Claude Fable 5.2 on the Agent Arena yet; it runs in Cowork on every paid Claude plan. In October, Anthropic's Claude Duet (October 21), a waitlisted research preview for US Pro and Max subscribers, joined the table rather than the podium, and Gemini 4 Argon reached the paid API and the Gemini app on Google AI Pro on October 13.
Among open models on the Agent Arena, Kimi K3.1 Pro, whose weights Moonshot published on November 2, is now the highest at 13th, ahead of Xiaomi's MiMo-V2.6-Pro at 18th and DeepSeek V4.1-Flash-1017 at 20th, while Kimi K3 sits 25th and Mistral's newly open Mistral Large 4 42nd. In October, OpenAI began rolling out texting for dots, by iMessage or RCS, to Pro subscribers in the US on October 28.
Press ⌥ Space on your Mac, ask, and keep working. Also on iPhone and iPad.
Free to start, 4.7★ across 27,000+ reviews.
Download Fello AIBest AI for Students
The best AI for students is GPT-6 Luna inside ChatGPT for general coursework and Gemini 3.6 Flash for STEM and multimodal study, with Qwen 3.7 Max as the API alternative for harder problem sets (200 free requests a day) and Claude Sonnet 5.5 as the free alternative for essay editing.
Most students don't need to pay. Since October 8, ChatGPT Free and Go have been moving from GPT-5.6 Luna to GPT-6 Luna, which Artificial Analysis scored 38.1 on Intelligence Index v4.3.2 against 37.3; the free plan has unlimited text chats, while file uploads and image tools stay capped. Claude Sonnet 5.5 is free on claude.ai, where free accounts can also pick Claude Haiku 5.5 (October 7), the Claude model that uses the least of your usage limit.
Gemini changed on October 9: the free Gemini app now runs on Auto, which sends most prompts to Flash-Lite, so choosing Gemini 3.6 Flash yourself needs Google AI Plus, though it stays free in Google AI Studio. For step-by-step working on the hardest math, GPT-6 Astra reports 97.6% on FrontierMath Tier 4 v2 but is not on ChatGPT's free tier; Qwen 3.7 Max is the value alternative at 97.1 HMMT 2026 February with API pricing at $1.25 / $3.75 on its current 50% promo ($2.50 / $7.50 list).
| Task | Best Model | Why | Free? | Alternative |
|---|---|---|---|---|
| Essays & coursework | GPT-6 Luna | ChatGPT's Free and Go model, rolling out since October 8; unlimited text chats | Yes | Claude Sonnet 5.5 (free Claude) |
| STEM problem-solving | GPT-6 Sol / Qwen 3.7 Max | ChatGPT's paid default; Astra reports 97.6% FrontierMath Tier 4 v2 / 97.1 HMMT 2026 Feb | Plus paid / Qwen API paid | Gemini 3.6 Flash (AI Plus since Oct 9) |
| Research & accuracy | Gemini 3.1 Pro | 98% ARC-AGI-1 at $0.52/task, native Google Search grounding | AI Pro (Gemini app) since Oct 9 | Claude Opus 5 |
| Writing editing | Claude Sonnet 5.5 | Free on claude.ai since September 28 and sixth among distinct models on Intelligence Index v5 at max effort; Claude Fable 5.1 is the quality leader | Yes (Claude free) | Claude Fable 5.1 |
| Multimodal study (PDFs, slides, images) | Gemini 3.6 Flash | 1M context via AI Studio; Gemini app needs AI Plus since Oct 9 | Free in AI Studio | NotebookLM (Google) |
Nothing changed for students since the last update. ChatGPT Free and Go run GPT-6 Luna, which OpenAI tuned on October 29 to give shorter answers with less flattery, Claude Sonnet 5.5 is still free on claude.ai, and the free Gemini app still runs on Auto, so picking Gemini 3.6 Flash in the app needs Google AI Plus.
In October, ChatGPT's free model moved from GPT-5.6 Luna to GPT-6 Luna from October 8, so GPT-6 Luna took the essays row, the free Gemini app moved to Auto on October 9, Anthropic released Claude Haiku 5.5 on October 7, which free Claude accounts can pick, and Gemini 4 Argon arrived on October 13 with Google AI Pro, so US students on Google's free year of AI Pro get it.
Best AI for Work & Professionals
The best AI for professional work is GPT-6 (ChatGPT's default since October 7, Sol on paid plans) for daily knowledge work, Claude Opus 5.5 for coding and high-stakes writing, and Gemini Spark for 24/7 agentic workflows.
Most professionals get the most out of running two paid subscriptions (ChatGPT Plus at $20/month plus Claude Pro at $20/month, total $40/month). Or keep one bill: Fello AI covers ChatGPT, Claude and Gemini for everyday work. For agentic work that runs while you sleep, Gemini Spark is the always-on cloud agent that lives inside Google Workspace, and since July 30 it reaches the $19.99/month Google AI Pro tier as well as Google AI Ultra; OpenAI dots, Grok Bot and Meta's Muse are the alternatives, covered in the agents section above.
| Use Case | Best Model | Key Stat | Price | Alternative |
|---|---|---|---|---|
| Daily knowledge work | GPT-6 | ChatGPT's default since October 7; GPT-6 Sol on Plus | $20/mo ChatGPT Plus | Claude Opus 5.5 |
| Legal, finance & tax research | Gemini 4 Argon | #1 HLAB 1.1 (27.1%), Harvey's legal-agent test inside the Vals Index; #2 Finance Agent v2.1 (67.7%); #2 AutomationBench-AA (77.5%), behind Gemini 4 Carbon | $19.99/mo Google AI Pro | Claude Opus 5.5 |
| ChatGPT Work & Codex | GPT-6.1 Sol | In ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu since September 29, at $2 / $10 in the API | $20/mo ChatGPT Plus | GPT-6 Luna for volume work |
| Coding (proprietary) | Claude Opus 5.5 | #1 Arena WebDev (1818) and #1 image-to-WebDev (1750), at $4 / $20 | $20/mo Claude Pro | Claude Fable 5.2 for the hardest work; Claude Sonnet 5.5 at half the token price |
| Coding (cost-effective) | Claude Sonnet 5.5 | #2 Arena WebDev (1774) and #3 Terminal-Bench 4.0 (63.6%); $1.64 an index task at high effort | $2 / $10 | Kimi K3.1 Pro; DeepSeek V4.1-Flash-1017 |
| Research & briefings | Gemini 3.1 Pro | 98% ARC-AGI-1 at $0.52/task, Google grounding | Google AI Pro / Ultra | Claude Opus 5.5 |
| Hard math, physics, finance modelling | GPT-6 Astra | #1 ARC-AGI-2 (95%), #2 LiveBench Reasoning; reports 97.6% FrontierMath Tier 4 v2 | $100/mo ChatGPT Pro | GPT-6.1 Sol; Qwen 3.7 Max |
| Always-on agent workflows | Gemini Spark | Always-on cloud agent inside Google Workspace | From $19.99/mo Google AI Pro | Claude Cowork |
| Live news, X-context creative | Grok 4.7 | 35.2 on Intelligence Index v5 + native X grounding; in the Grok app since October 3 | $30/mo Premium (X Subscription) | Gemini 3.1 Pro |
| All-in-one consolidation | Fello AI | ChatGPT + Claude + Gemini + Grok + DeepSeek | From $11.99/mo in the US (prices vary by country) | Pay each vendor separately |
The daily knowledge work row stays with GPT-6, ChatGPT's default since October 7, and no row changed hands since the last update: Gemini 4 Argon keeps the legal, finance and tax row, because Gemini 4 Carbon, generally available since November 10, leads only AutomationBench-AA and sits eighth on HLAB 1.1 at 17.5% and seventh on Finance Agent v2.1 at 67.16%, the two boards the row is decided on, and Claude Opus 5.5 keeps the proprietary coding row, now at 1818 on Arena WebDev, with Claude Fable 5.2, released November 9, as the alternative for the hardest work. In October, Vals re-ran its index on October 27 with Claude Sonnet 5.5 first and Carbon second, Argon took the new legal, finance and tax row on October 13 on HLAB 1.1 and Finance Agent v2.1, Kimi K3.1 Pro took Terminal-Bench 4.0 from Sonnet 5.5, and the coding rows moved to FrontierCode 2.0 from Intelligence Index v5.
AI Model Pricing in November 2026
From $0 free tiers to $199.99/month Google AI Ultra. The price that matters most on this table is Claude Fable 5.2 at $8 / $40, down from $10 / $50 for Fable 5.1: the highest-scoring model on Artificial Analysis's Intelligence Index v5, at 64.6, costs twice Claude Opus 5.5's $4 / $20, and at max effort an index task costs $26.54 against Opus 5.5's $10.04. At high effort it scores 61.8 for $9.08, above Opus 5.5 at max for slightly less, so the effort setting decides whether it is the expensive option. Its new fast mode costs $16 / $80, and Claude Mythos 5.2 bills identically for verified organizations. GPT-6 Astra lists at $10 / $50 and scores 56.9. Google's Gemini 4 Carbon scores 60.0 at the same $2 / $10 as Argon until November 30, and it has been on the paid API since November 10.
The busiest rate is now $2 / $10, shared by Claude Sonnet 5.5, GPT-6 Sol, GPT-6.1 Sol and Moonshot's Kimi K3.1 Pro, which scores 46.5 at $3.68 an index task. GPT-6.1 Sol is still the cheapest route past 50 on the index, at $0.59 an index task for 50.3 at high effort, or $1.27 for 54.6 at max, and OpenAI's GPT-6.1 Terra (October 26) halves the token price to $1 / $5 for 51.3 at $0.96. With Sonnet 5.5 the effort setting decides the bill: an index task costs $1.64 at high effort and $10.02 at max, level with Opus 5.5's $10.04, and Anthropic halved its cache reads to $0.10 on October 7. Google's Gemini 4 Argon also starts at an introductory $2 / $10 ($4 / $20 standard), until November 30, and since October 13 anyone on the paid API can buy it.
Token price and task cost can drift apart. Grok 4.7 lists at $2 / $6, yet an index task costs $6.18 for a score of 35.2, because it writes so much output. Mistral Large 4, whose weights Mistral published on October 29, now lists at $0.70 / $2.10, half its preview rate, and an index task costs $2.06 for a score of 26.1. Meta's Muse Spark 1.3 at $1.25 / $4.25 was the frontier value pick on v4.3.2, at $1.60 an index task for 48.1, and Artificial Analysis has not yet scored it on v5.
At the cheap end, Meta's Muse Ember (November 9) is now the cheapest model per task on this page: $0.10 / $0.50, or $0.14 an index task for 36.0 at max effort. GPT-6 Luna, at the same $0.10 / $0.50, costs $0.15 an index task for 27.9. Claude Haiku 5.5 (October 7) matches that token price on prompts up to 100,000 tokens and scores 36.5 to Luna's 27.9, but it writes so much more that an index task costs $0.41 at max effort, or $0.25 for 34.6 at xhigh; above 100,000 tokens it rises to $0.50 / $2.50. Among open models sold by the token, DeepSeek V4.1-Flash-1017, the build DeepSeek's API has served since October 17, is the pick at $0.15 / $0.60 off-peak and $0.30 / $1.20 at peak ($0.60 an index task for 32.6, second among open models on v5 behind Kimi K3.1 Pro), while GLM 5.3 Flash at $0.15 / $0.50 now scores 27.8 on v5, at $0.58 an index task. Xiaomi's MiMo-V2.6-Pro is cheaper per task at $0.26 for 29.9, but only if you host its 1.02 trillion parameters yourself. For a deeper breakdown see our full AI Pricing Comparison Guide.
| Model | Input (per 1M) | Output (per 1M) | Context | Free Access? |
|---|---|---|---|---|
| GPT-6 Astra | $10.00 | $50.00 | 1,050,000 (922,000 max input) | In Chat on Pro, Business and Enterprise since September 4; Plus gets it in ChatGPT Work and Codex rather than Chat; no free tier; live in the API |
| GPT-6 Sol | $2.00 | $10.00 | 1,050,000 (922,000 max input) | ChatGPT Chat on Plus, Pro, Business and Enterprise since October 7; Work & Codex; API |
| GPT-6.1 Sol | $2.00 | $10.00 | 1,050,000 (922,000 max input) | ChatGPT Work & Codex on Plus, Pro, Business, Enterprise and Edu since September 29; API; not in Chat; cache reads $0.10 |
| GPT-6.1 Terra | $1.00 | $5.00 | 1,050,000 (922,000 max input) | ChatGPT Work & Codex on Plus, Pro, Business, Enterprise and Edu since October 26; API; not in Chat; cache reads $0.05 |
| GPT-6 Luna | $0.10 | $0.50 | 1,050,000 (922,000 max input) | ChatGPT Chat on Free and Go, rolling out since October 8; Work & Codex on paid plans; API |
| GPT-5.5 | $5.00 | $30.00 | 1M (400K in Codex) | ChatGPT Free; API paid |
| GPT-5.5 Pro | $30.00 | $180.00 | 1M | ChatGPT Pro from $100/mo ($200 and $500 higher-usage tiers) |
| GPT-5.6 Sol | $4.00 | $20.00 | Not published | Live in ChatGPT, Codex & API (July 9); promotional rate through at least November 21, 2026 |
| GPT-5.6 Terra | $2.00 | $12.00 | Not published | Live in ChatGPT, Codex & API (July 9) |
| GPT-5.6 Luna | $0.20 | $1.20 | Not published | Live in ChatGPT, Codex & API (July 9) |
| Claude Opus 5.5 | $4.00 | $20.00 | 1M | Claude Pro/Max/Team; API paid; cache reads $0.20 |
| Claude Opus 5 | $5.00 | $25.00 | 1M | Legacy model at Anthropic; Pro/Max/API |
| Claude Opus 4.8 | $5.00 | $25.00 | 1M | Legacy model at Anthropic; Pro/Max/API |
| Claude Fable 5.2 | $8.00 | $40.00 | 1M | Every paid Claude plan: included up to ~50% of weekly limits on Max, Ultra and premium Team and Enterprise seats; Pro and standard seats via usage credits (included until November 16); cache reads $0.20; fast mode $16 / $80. Mythos 5.2 bills identically, verified organizations only |
| Claude Fable 5.1 | $10.00 | $50.00 | 1M | Legacy model since November 9; cache reads $0.25 |
| Claude Fable 5 | $10.00 | $50.00 | 1M | Permanent in Max/Team Premium (~50% of usage limits); Pro/Team Standard via credits |
| Claude Sonnet 5.5 | $2.00 | $10.00 | 1M | Free on claude.ai; Pro/Max/Team; API paid; cache reads $0.10 since October 7 |
| Claude Sonnet 5 | $2.00 | $10.00 | 1M | Legacy model since September 28; the fallback for flagged cyber requests; API paid |
| Claude Sonnet 4.6 | $3.00 | $15.00 | 1M | API paid (superseded by Sonnet 5) |
| Claude Haiku 5.5 | $0.10 (≤100K) / $0.50 (>100K) | $0.50 (≤100K) / $2.50 (>100K) | 1M | Free, Pro, Max, Team and Enterprise on Claude.ai (web, iOS, Android); API paid; cache reads $0.01 |
| Gemini 3.1 Pro | $2.00 (≤200K) / $4.00 (>200K) | $12.00 (≤200K) / $18.00 (>200K) | 1M | Gemini app on Google AI Pro and above since Oct 9; API paid |
| Gemini 4 Argon | $2.00 intro to Nov 30 / $4.00 standard ($0.10 cached, intro) | $10.00 intro / $20.00 standard | 1M, shared by input and output | Paid API since October 13; Gemini app on Google AI Pro and Ultra |
| Gemini 4 Carbon | $2.00 until Nov 30 / $4.00 from Dec 1 | $10.00 until Nov 30 / $20.00 from Dec 1 | Not published | Generally available on the paid API since November 10; in the Gemini app on Google AI Ultra only |
| Gemini 3.8 Flash | $0.75 intro / $1.50 from Jan 1, 2027 | $3.75 intro / $7.50 from Jan 1, 2027 | 1M | AI Studio, Antigravity, Gemini Enterprise + paid API; in the Gemini app reported for AI Pro/Ultra |
| Gemini 3.7 Flash | $0.75 intro / $1.50 from Jan 1, 2027 | $3.75 intro / $7.50 from Jan 1, 2027 | 1M | AI Studio, Android Studio, Antigravity + paid API; in the Gemini app via Spark only (AI Pro/Ultra, excludes EEA/UK/CH/Nigeria) |
| Gemini 3.6 Flash | $0.75 intro / $1.50 from Jan 1, 2027 | $3.75 intro / $7.50 from Jan 1, 2027 | 1M | Gemini app on AI Plus and above since Oct 9; AI Studio; free API tier + paid API |
| Gemini 3.5 Flash-Lite | $0.30 | $2.50 | 1M | AI Studio; free API tier + paid API |
| Qwen3.8-Max-0902 | $2.00 ($0.17 explicit cache hit, $0.25 implicit) | $6.00 | 1M (128K output) | QwenCloud API only; no consumer chat front-end; open weights only as the text-only Qwen3.8-2.4T-A95B it is built on |
| Qwen 3.7 Max | $1.25 promo / $2.50 list | $3.75 promo / $7.50 list | 1M | 200 free requests/day; API paid beyond that |
| MiniMax M3 | $0.30 (50% off $0.60) | $1.20 (≤512K) | 1M | Open weights; hosting costs apply |
| LongCat-2.0 | Provider-dependent | Provider-dependent | 1M | Open weights (MIT); hosting costs apply |
| NVIDIA Nemotron 3 Ultra | Provider-dependent | Provider-dependent | 1M | Open weights (OpenMDW); hosting costs apply |
| Qwen 3.5 (open-weight) | Self-host / Together | Self-host / Together | 1M | Open weights; hosting costs apply |
| Nex-N2-Pro | Self-host / providers | Self-host / providers | 1M | Open weights (Apache 2.0); hosting costs apply |
| Rio 3.5 Open 397B | Self-host / providers | Self-host / providers | 1M | Open weights (MIT); hosting costs apply |
| Grok 4.3 | $1.25 | $2.50 | 1M | Free consumer plan; API paid |
| Grok 4.6 | $2.00 (<200K) / $4.00 (≥200K) | $6.00 (<200K) / $12.00 (≥200K) | 500K | SpaceXAI API, Cursor, Grok Build, OpenRouter, Vercel, Cloudflare |
| Grok 4.7 | $2.00 (<200K) / $4.00 (≥200K) | $6.00 (<200K) / $12.00 (≥200K) | 500K | SpaceXAI API, Cursor, Grok Build, harnesses and cloud routers; Grok app Expert mode since October 3 |
| Muse Spark 1.3 | $1.25 ($0.10 Contributor tier) | $4.25 ($0.20 Contributor tier) | 1M | Paid API; Muse Code, Meta Model API and OpenRouter; Contributor tier trades training rights for a ~92% discount |
| Muse Wonder | $4.00 | $20.00 | 1M | Paid API; Meta Model API, Muse Code and OpenRouter; Muse Wonder Full is only in Muse Code, on the $50 and $200 plans |
| Muse Ember | $0.10 (≤200K) / $0.30 (>200K) | $0.50 (≤200K) / $1.50 (>200K) | Not published | Muse Code and the Meta Model API since November 9; no consumer app, no open weights; cached input $0.004 |
| Kimi K3 | $3.00 ($0.30 cache-hit) | $15.00 | 1M | Free basic tier in the Kimi app; open weights on Hugging Face (Kimi K3 License) |
| Kimi K3.1 Pro | $2.00 ($0.20 cache-hit) | $10.00 | 1M | Kimi.com, Kimi apps, Kimi Work, Kimi Code and the Kimi API since October 23; open weights on Hugging Face since November 2 (Kimi K3.1-Pro License); Highspeed variant at twice the price |
| Mistral Large 4 | $0.70 ($0.07 cached) | $2.10 | 1M | Mistral API, Mistral Studio, OpenRouter and Microsoft Foundry; open weights on Hugging Face since October 29 under the Mistral Ecosystem Open-Weight License (modified MIT) |
| Gemini Omni 1.2 Flash (video) | $1.50 | $12.00 (video output) | 3-15 second clips | Gemini app / Flow / Vids; AI Studio + API |
| DeepSeek V4-Pro | $0.66 off-peak / $1.32 peak ($0.022 cache-hit off-peak) | $1.98 off-peak / $3.96 peak | 1M | DeepSeek Chat free; API paid, peak/off-peak tiers since August 16 |
| DeepSeek V4.1-Flash-1017 | $0.15 off-peak / $0.30 peak ($0.003 cache-hit off-peak) | $0.60 off-peak / $1.20 peak | 1M | DeepSeek Chat free; API paid, peak/off-peak tiers; 1017 build since October 17 at the same prices |
| Kimi K2.7 Code | Provider-dependent | Provider-dependent | 256K | Open weights; hosting costs apply |
| GLM-5.2 | Provider-dependent | Provider-dependent | 1M | Open weights; hosting costs apply |
| GLM 5.3 | $1.40 ($0.26 cache-hit) | $4.40 | 1M | Z.ai API, GLM Coding Plan and ZCode; weights on Hugging Face since August 28 under the bespoke GLM 5.3 License |
| GLM 5.3 Flash | $0.15 ($0.03 cache-hit) | $0.50 | 1M | Z.ai API, GLM Coding Plan, OpenRouter; MIT weights on Hugging Face; the launch promotion ended September 9 |
| Tencent Hy4 Preview | $0.834 ($0.042 cache-hit) | $2.501 | 1M+ | Open weights (Apache 2.0); Tencent Cloud TokenHub, OpenRouter, WorkBuddy, CodeBuddy |
| Qwen3.8-Flash-Next | Provider-dependent | Provider-dependent | 262K (to 1M) | Open weights (Qwen Community License 1.0); hosting costs apply |
| MiMo-V2.6-Pro | $0.435 | $0.87 | 1M | Open weights (MIT); hosting costs apply |
| ERNIE 5.1 | China-region pricing | China-region pricing | 256K | Baidu free tier |
| Gemini Spark (agent) | Not API-priced | Not API-priced | 1M (Gemini base) | Google AI Pro $19.99, AI Ultra $99.99 or $199.99/mo |
| Fello AI (aggregator) | Routed via app | Routed via app | Model-dependent | From $11.99/mo in the US (prices vary by country), free tier available |
The GPT-5.5 and GPT-5.5 Pro rates above are short-context prices; OpenAI no longer publishes the specific long-context figures. The GPT-5.6 tiers are billed at 2x input and 1.5x output once a prompt passes 272K input tokens, which puts long-context Terra at $4 / $18 and Luna at $0.40 / $1.80. The GPT-6 tiers carry the same surcharge past 272K input tokens, applied to the whole request, and their cached input reads are 90% off: $1.00 for Astra, $0.20 for GPT-6 Sol and $0.01 for Luna, with GPT-6.1 Sol at 95% off, $0.10.
Grok 4.6 works the same way but bites harder: at 200,000 prompt tokens and above, SpaceXAI re-bills the entire request at the higher rate rather than only the overage, so crossing the line by a thousand tokens doubles the cost of the whole call. DeepSeek is the other rate card that needs reading twice: since August 16 both V4 models bill at peak rates from 01:00 to 04:00 and from 06:00 to 10:00 UTC on weekdays, and at exactly half that in every other hour, so the same job can cost twice as much depending on when you run it.
The cheapest way through this table is often one subscription instead of three. Fello AI puts GPT, Claude, Gemini, Grok, Perplexity and more in a single app for Mac, iPhone and iPad, from $11.99 a month in the US (prices vary by country).
Best Open-Weight Models in November 2026
The best open-weight model in November 2026 is Moonshot's Kimi K3.1 Pro, whose weights went up on Hugging Face on November 2: 2.8 trillion parameters with 104 billion active and a 1M-token context, under Moonshot's own Kimi K3.1-Pro License. Artificial Analysis scores it 46.5 on its new Intelligence Index v5, 13.9 points above the next open model, DeepSeek V4.1-Flash-1017 at 32.6, and it is first overall on Terminal-Bench 4.0 at 67.2%. It is also the highest-placed open model on Arena's WebDev board, tenth, and on the Agent Arena, 13th. It costs $2 / $10 per 1M tokens on the Kimi API, or $3.68 an index task, and the download is about 1.56 TB, so it is a data-centre model, not one for a workstation.
DeepSeek's V4.1-Flash-1017, released on October 17 under MIT, held this spot until November 2 and is still the best open value sold by the token: 32.6 on v5 at $0.60 an index task, $0.15 / $0.60 per 1M tokens off-peak, 552 billion parameters with 8 billion active at prefill and 16 billion at decode, and 16th on Arena WebDev. Xiaomi's MiMo-V2.6-Pro, published on September 21 under plain MIT, is the cheapest strong open model per task at $0.26, and GLM-5.3 reaches 41.9% on Terminal-Bench 4.0, though on a bespoke Z.ai licence rather than MIT.
One October release sits just outside this ranking. MiniMax-M3.1-Flash, whose weights are on Hugging Face under MiniMax's community licence, scores 29.1 from 428B total and 23B active parameters at $0.49 an index task, a little cheaper per task than DeepSeek V4.1-Flash-1017 but 3.5 points lower.
For teams running their own weights, the practical pick is still GLM 5.3 Flash (Z.ai, MIT): 27.8 on v5, or 41.8 on v4.3.2, from 320B total and 18B active. It is a separate model on a newly trained base, not a trimmed GLM 5.3, which is why it ships under MIT. A 1.02-trillion-parameter model is not a workstation proposition, and the K3 download alone is about 1.56 TB. DeepSeek-V4.1-Flash-1017 replaced the September V4.1-Flash build on October 17 at unchanged prices: MIT, natively multimodal, and $0.15 / $0.60 off-peak.
Four more are worth watching. Mistral Large 4, whose weights Mistral published on October 29 under its own modified-MIT licence, is the highest-scoring open model from the US or Europe, at 26.1 on v5 for $2.06 an index task, but that is only eighth among open models, and a business that sells it as a hosted service needs a separate commercial licence from Mistral. Alibaba's Qwen3.8-Flash-Next, 125B with only 6B active, previews the Qwen4 architecture and scored 39.8 on v4.3.2. Tencent's Hy4 Preview, 770B total and 49B active under Apache 2.0, has no Artificial Analysis rating but sits 24th on Arena WebDev. Meta's Muse Spark Open, released October 15 at 816B total and 42B active, scores 24.9 on v5 under Meta's own Muse Model License rather than an open-source licence.
| Model | Best For | Key Benchmark | Context / License | Where To Run |
|---|---|---|---|---|
| Kimi K3.1 Pro | Best open model | II 46.5 (v5.0), the highest open score by 13.9 points, at $3.68 an index task; #1 Terminal-Bench 4.0 (67.2%); top open model on Arena WebDev (#10) and the Agent Arena (#13); 2.8T/104B active | 1M / Kimi K3.1-Pro License | Hugging Face (96 shards, 1.56 TB), Kimi API ($2/$10) |
| DeepSeek V4.1-Flash-1017 | Best open value sold by the token | II 32.6 (v5.0), second among open models, at $0.60 an index task; #16 Arena WebDev, #20 on the Agent Arena; 552B, 8B active at prefill and 16B at decode | 1M / MIT | DeepSeek API ($0.15/$0.60 off-peak, $0.30/$1.20 peak), local |
| Kimi K3 | Kimi K3.1 Pro's predecessor, same 2.8T size | II 43.6 (v4.3.2); #17 Arena WebDev and #25 Agent Arena; 2.8T/104B active | 1M / Kimi K3 License | Hugging Face (96 shards, 1.56 TB), Moonshot API, providers |
| GLM 5.3 Flash | Best open model most teams can actually run | II 27.8 (v5.0) at $0.58 an index task, 41.8 on v4.3.2, the smallest of the leading open models; 320B/18B active, natively multimodal | 1M / MIT | Hugging Face, Z.ai API ($0.15/$0.50), OpenRouter |
| GLM-5.2 | Fallback with an independent track record | II 33.7 (v4.3.2); #35 Arena WebDev (1608), #28 Agent Arena; 744B/40B active | 1M / MIT | Z.ai, Hugging Face, OpenRouter |
| GLM 5.3 | Open since August 28, but on a bespoke licence | II 31.7 (v5.0), third among open models; same 744B base as GLM-5.2, post-trained only, published checkpoint counted at 753B; CyberGym 84.5% and Terminal-Bench 3.0 28.3 (vendor figures) | 1M / GLM 5.3 License (MaaS above $10B revenue needs a Z.ai security review) | Hugging Face, Z.ai API ($1.40/$4.40), GLM Coding Plan, ZCode |
| MiMo-V2.6-Pro | Cheapest strong open model per task | II 29.9 (v5.0), fourth among open models, at $0.26 an index task; 34.8% Terminal-Bench 4.0; GDPval-AA v2.1 1689; #18 Agent Arena, second among open models there after Kimi K3.1 Pro; 1.02T/42B active | 1M / MIT | Hugging Face (XiaomiMiMo), providers, self-host |
| MiniMax-M3.1-Flash | Cheap open coder and agent | II 29.1 (v5.0) at $0.49 an index task; 428B/23B active, text, image and video input; #41 Arena WebDev (1582) | 1M / MiniMax Community License | Hugging Face, MiniMax, self-host |
| Mistral Large 4 | Highest-scoring open model from the US or Europe | II 26.1 (v5.0) at $2.06 an index task, eighth among open models; about 1T total and 49B active, natively multimodal; #45 Arena WebDev (1562) | 1M / Mistral Ecosystem Open-Weight License (modified MIT) | Hugging Face, Mistral API ($0.70/$2.10), OpenRouter, Microsoft Foundry |
| Tencent Hy4 Preview | Largest permissively licensed model yet | 770B/49B active; 2.99/4.00 on Tencent's own 203-task expert panel vs Kimi K3 2.94 and GLM 5.3 2.92 (vendor); no Artificial Analysis rating; #24 Arena WebDev (1631), #26 Agent Arena | 1M+ / Apache 2.0 | Hugging Face (BF16 and FP8), Tencent Cloud TokenHub, OpenRouter |
| Qwen3.8-Flash-Next | Preview of the Qwen4 architecture | 125B total plus a 51B N-gram embedding, 6B active; 91.7 GPQA Diamond and 62.5 SWE-Bench Pro (vendor); II 39.8 (v4.3.2); #26 Arena WebDev (1624) | 262K to 1M / Qwen Community License 1.0 | Hugging Face, providers, self-host |
| LongCat-2.0 | Frontier open coder trained on Chinese chips | II 33 (v4.1); 59.5% SWE-Bench Pro (vendor), 1.6T/~48B active | 1M / MIT | Hugging Face, GitHub, OpenRouter |
| MiniMax M3 | Cheap long-context multimodal | II 18.9 (v5.0; 29.2 on v4.3.2), 59% SWE-Bench Pro, multimodal | 1M / MiniMax Community License | Hugging Face, API $0.30/1M (50% off) |
| Nex-N2-Pro | Strongest open coding score | II 41 (v4.1); 80.8 SWE-Bench Verified, 397B/17B active | Qwen-based / Apache 2.0 | Hugging Face, providers, self-host |
| Kimi K2.7 Code | Strongest commercially-licensed open coder | +21.8% on Kimi Code Bench v2 vs K2.6 (vendor); 1T/32B active | 256K / Modified MIT | Hugging Face, DeepInfra, providers |
| DeepSeek V4-Pro | Agentic real-world work | II 44 (v4.1), 1.6T/49B active | 1M / MIT | DeepSeek API ($0.66/$1.98 off-peak, $1.32/$3.96 peak since August 16), local |
| Hy3 | Newest permissive-licence entrant | II 41 (v4.1); #66 on Arena WebDev (1506) | Apache 2.0 | Hugging Face, providers, self-host |
| Inkling | Thinking Machines' first open model | II 41 (v4.1), agentic 32.3, released July 15 | Open weights | Hugging Face, providers, self-host |
| Inkling Small | Same family at a quarter the size | II 40 (v4.1), agentic 30.8; 276B/12B active, text, image and audio in | Apache 2.0 | Hugging Face (BF16 and NVFP4), providers, self-host |
| NVIDIA Nemotron 3 Ultra | NVIDIA-tuned, fully permissive license | II 38 (v4.1), 65-70.4 SWE-Bench Verified, 550B/55B active | 1M / OpenMDW | OpenRouter, Hugging Face, AWS (8x B200 self-host) |
| Qwen 3.5 (397B / 17B active) | Multimodal, fast decode | 88.4 GPQA, 91.3 AIME 2026, 83.6 LiveCodeBench v6 | 1M / open | Together, OpenRouter, local |
| Qwen3.6-35B-A3B | Efficient open agentic coder (3B active) | 86.0 GPQA Diamond, 92.7 AIME 2026, 35B/3B active | 262K (to 1M YaRN) / Apache 2.0 | Hugging Face, OpenRouter, local |
| Qwen3.6-27B | Laptop-runnable dense coder | 87.8 GPQA Diamond, dense 27B, multimodal | 256K / Apache 2.0 | Local Mac/PC, Hugging Face, OpenRouter |
| Rio 3.5 Open 397B | Qwen 3.5 fine-tune, multilingual reasoning | 70.8 Terminal-Bench 2.1 (first-party), beats Qwen 3.7 Plus on 4/5 | 397B/17B active / MIT | Hugging Face, providers, self-host |
| Llama 4 Maverick | Meta-line flagship | 17B active / 400B total params | Llama 4 license | Meta cloud, Hugging Face, local |
| NVIDIA Nemotron 3 Nano Omni | Edge / low-power | Multimodal, very small footprint | Compact / open | Local, NVIDIA tool |
Licensing matters as much as raw score here, and August widened the gap between the two groups. Kimi K2.7 (Modified MIT), DeepSeek V4 (MIT), GLM 5.3 Flash (MIT), MiMo-V2.6-Pro (MIT), GLM-5.2 (MIT), LongCat-2.0 (MIT), Hy3 (Apache 2.0), Hy4 Preview (Apache 2.0), Nex-N2-Pro (Apache 2.0) and Nemotron 3 Ultra (OpenMDW) all clearly allow commercial use with no revenue test.
The rest carry conditions worth reading before you build on them: MiniMax M3, MiniMax-M3.1-Flash and MiniMax H3 ship under their own community licences, H3 additionally requiring an application from the US, EU, UK and South Korea; Mistral Large 4's modified MIT licence requires a separate commercial licence from Mistral for anyone selling it as a hosted service, though internal use is exempt; Kimi K3 and Kimi K3.1 Pro sit on Moonshot's own licences, which require a separate agreement from any business with more than $20 million in revenue that resells them as a service; GLM 5.3 requires a Z.ai security review of any model-as-a-service operator above $10 billion in revenue; and the Qwen Community License 1.0 on Qwen3.8-Flash-Next requires a separate licence from Qwen to run a model-as-a-service or an AI work-assistant business, though internal use is exempt. No open model is first on any Arena board we track any more: Claude Opus 5 took the Agent board in July, the last one an open model led.
Benchmarks, Prices, and Hands-On Use
Every ranking on this page combines three inputs: public benchmarks from seven independent houses (Artificial Analysis, Arena formerly LMArena, Scale SEAL, LiveBench (version 2.0 since November 11, with Language and Mathematics figures from the archived 1.0), EQ-Bench, ARC Prize and the official Terminal-Bench 4.0 board, covering the Intelligence Index, GPQA Diamond, ARC-AGI-1 through 3, Humanity's Last Exam, GDPval-AA, FrontierMath, HMMT, MCP Atlas, SWE Atlas and the Remote Labor Index), published API and subscription pricing from each vendor's official pricing page, and hands-on use by the FelloAI editorial team running real prompts across the same task on every model. We re-fetch official pricing and benchmark sources before every monthly update.
Benchmarks are weighted to the use case: SWE-bench and Terminal-Bench drive coding, GPQA Diamond and ARC-AGI-2 drive accuracy, GDPval-AA (Artificial Analysis's professional-deliverables benchmark) informs professional-task quality while writing style is judged primarily by hands-on testing, FrontierMath and HMMT drive problem-solving. We disclose when a benchmark is vendor-reported but not independently verified, and we strip any claim we cannot reproduce against a live source. We rank on measured scores: a category crown moves as soon as a model out-scores the holder on the boards that define that category, without waiting for the vote-based boards to catch up, and a model that is not yet widely available is flagged on its card rather than denied one. We re-check ranks as well as figures, because a score can stay correct while the board around it moves. When a model goes through a major upgrade between updates, we re-rank the category and add a "What changed this month" line at the bottom of the deep-dive.
Ask ChatGPT, Claude and Gemini and keep the best answer. No three bills.
Fello AI runs on Mac, iPhone and iPad, free to start.
Download Fello AI
