Grok 4.3 Review: File Generation, 1M Context, and a 40% API Price Cut

xAI just rolled Grok 4.3 out to the full API on April 30, 2026, two weeks after a $300/month beta locked behind SuperGrok Heavy. The bigger story is the price reset. Grok 4.3 now runs at $1.25 per million input tokens and $2.50 per million output tokens, an input cut of roughly 40% and an output cut closer to 60% versus Grok 4.20. It scored 53 on the Artificial Analysis Intelligence Index as that index stood in May 2026, slotting just above Claude Sonnet 4.6 and well below GPT-5.5, Claude Opus 4.7, and Gemini 3.1 Pro Preview.

This Grok 4.3 review covers what actually changed since the April beta, where Grok now stands against the other frontier models (including Alibaba’s new Qwen3.7-Max), and the honest answer to the question every comparison-shopper has right now: should you pay $30 for SuperGrok, $300 for SuperGrok Heavy, or skip it altogether and use the API. We will also walk through how Mac and iPhone users can run Grok 4.3 alongside ChatGPT, Claude, Gemini, and DeepSeek without juggling five separate subscriptions. For the broader iPhone lineup beyond Grok, see our updated best AI apps for iPhone ranking.

Update, September 25, 2026: Grok 4.3 is two generations old. xAI’s API flagship is now Grok 4.7, which scores 46.3 on Artificial Analysis Intelligence Index v4.3.2 against 24.9 for Grok 4.3, and the index has been rescaled twice since this review was written, so the 53 below is not comparable with any number you read today. The Grok app sells modes rather than model numbers, and the two modes that name a model, Expert and Heavy, both say Grok 4.6. Grok 4.3 itself is still on xAI’s price list at the same rates, and Artificial Analysis now marks it deprecated. Read the rest of this review as the record of what Grok 4.3 brought in May 2026.

The Key Takeaways

• Grok 4.3 went GA on April 30, 2026 after an April 17 beta on SuperGrok Heavy.
• API pricing is $1.25 input / $2.50 output per million tokens, ~40% / ~60% cheaper than Grok 4.20.
• Native video input (max 5 minutes, 1080p, mp4/mov/webm) and native PDF, PPTX, and XLSX generation in chat.
• 1M-token context window on the API, and 53 on the Artificial Analysis Intelligence Index as it was scored in May 2026.
• API access is open to all developers and Grok 4.3 is still on xAI’s price list; the staged consumer rollout finished long ago, and the Grok app now sells modes rather than model numbers.

What Is Grok 4.3?

Grok 4.3 was xAI’s flagship reasoning model when it launched, released in beta on April 17, 2026 and rolled out via API on April 30, 2026 per the official xAI models documentation. It scored 53 on the Artificial Analysis Intelligence Index of the day, brought file generation (PDF, PPTX, XLSX) into the chat, and runs at $1.25 per million input tokens with a 1 million token context window. It was the cheapest frontier-tier model xAI had shipped at the time.

The model accepts text and image inputs and returns text, which is what xAI’s own model registry lists for grok-4.3 today. Reasoning effort is selectable from none through xhigh, with low as the default. Knowledge cutoff is December 2025.

What’s New in Grok 4.3 vs Grok 4.20

Grok 4.20 was a strong reasoning model that few people used because of how it was priced. Grok 4.3 fixes that and adds three serious capabilities on top.

The video input claim

This was the headline feature in launch week. Coverage at the time said you could drop a video file, up to 5 minutes at 1080p or lower, in mp4, mov, or webm, straight into Grok 4.3 and ask questions about it, billed as image tokens after frame extraction. It does not survive a check against the vendor. xAI’s own model registry lists grok-4.3 with text and image inputs and text output only, xAI has published no Grok 4.3 announcement that is still reachable, and the 5-minute, 1080p specification appears in no xAI document we can find. Treat video input on this model as unconfirmed rather than as a capability to plan around.

Native file output

Grok 4.3 can now generate downloadable PDFs, fully formatted PowerPoint decks, and populated Excel spreadsheets directly inside the chat. Early hands-on reports from independent testers describe outputs that are clean enough to actually send, not the rough first drafts that earlier “export” features produced.

1M-token context on the API

Grok 4.20 already had a long context, but Grok 4.3 standardises 1 million tokens across the API with no output token cap. That is roughly 2,000 pages of text in one prompt, which makes long-document analysis and agentic workflows substantially easier.

Pricing reset

This is the part most coverage from April missed, as VentureBeat noted in its launch coverage. Grok 4.3 launched on the beta locked to SuperGrok Heavy at $300/month. After the API GA on April 30, API pricing dropped to $1.25 input and $2.50 output per million tokens. xAI then rolled Grok 4.3 chat access out to standard SuperGrok and X Premium+ subscribers in stages. That rollout is history now: the Grok app no longer names Grok 4.3 anywhere, and its Expert and Heavy modes both say Grok 4.6. The API rates have not moved, and they still undercut OpenAI’s current mid flagship, GPT-6.1 Sol at $2 / $10. Our breakdown of is Grok free covers exactly what the no-cost X tier still gives you today.

There are also smaller updates worth noting: an Auto / Fast / Expert mode selector that controls how hard Grok thinks, the Grok Imagine agent for creative video and image work (including the R-rated Spicy Mode), and improved voice cloning. We break down what the Imagine agent can actually produce, from prompt to finished clip, in our Grok Imagine video generation walkthrough. The 16-Agent Heavy parallel orchestration system remains exclusive to the $300 SuperGrok Heavy plan.

Grok 4.3 Pricing: SuperGrok, SuperGrok Heavy, and API

For a full breakdown of every plan, including X Premium+, Imagine credits, and team pricing, see our Grok pricing guide. The consumer ladder has grown since this review: the App Store listing for the Grok app now publishes SuperGrok Lite at $10 and SuperGrok Plus at $100 alongside the $30 and $300 tiers.

Grok 4.3 Benchmarks: How It Compares

The Artificial Analysis Intelligence Index was the cleanest cross-model snapshot at the time. Grok 4.3 scored 53, a four-point gain over Grok 4.20, which put it above Claude Sonnet 4.6 and Muse Spark and below the leaderboard of the day. The table below is a May 2026 snapshot on the index version of that month. Artificial Analysis has rescaled the index twice since, so none of these numbers is comparable with a score you read today: on v4.3.2 Grok 4.3 reads 24.9 and is marked deprecated.

The takeaway at the time: Grok 4.3 was not the smartest model on the board, but at $1.25 / $2.50 per million tokens it was the cheapest one with a 1M context and Office-style file output, and for agentic workflows that hammer on long documents the price-to-capability ratio was hard to beat. Both rivals in that table have since been superseded, so for complex coding, scientific reasoning, or deep research today the comparison to run is against GPT-6.1 Sol and Claude Opus 5.5.

How to Use Grok 4.3 on Mac and iPhone

xAI does ship an official Grok app for iPhone and iPad, published by X Corp. on the App Store, but there is still no native Mac app. You have three ways to reach Grok from an Apple device, and none of them lets you pick Grok 4.3 by name.

The first option is grok.com in your browser, which works fine on Safari, Chrome, or Arc but lives behind another browser tab. The second is the X app or X.com, which gives you Grok inside the social-network UI and is mostly useful if you already pay for X Premium+. The third option, and the one we use day to day, is Fello AI for Mac and iPhone, which gives you Grok alongside Claude, ChatGPT, Gemini, and DeepSeek under one $9.99/month subscription. No model lock-in, no app switching, and one chat history across providers.

Mac users running M-series chips can also call the Grok 4.3 API directly from local tools like Raycast or terminal scripts, which is the cheapest path if your usage is bursty.

Who Should Actually Pay for Grok 4.3

The honest verdict depends on what you do with it.

That question has changed shape. You can no longer buy Grok 4.3 chat access at all, because xAI sells consumer access by mode rather than by model number: Fast is the free one, and Auto, Expert and Heavy require a paid plan. The two that name a model both say Grok 4.6. If you specifically want Grok 4.3, it is now an API model.

If you build agents, do API integrations, or run large document workloads, the API at $1.25/M input is still the cheapest way to reach a 1M-token context inside xAI’s own line, and it remains competitive with DeepSeek on price.

If you mostly want a chatbot that handles your daily writing, research, and image work, you are better served by a multi-model setup. Comparing Grok against ChatGPT shows where each model is strong: Grok wins on real-time information, X-data-aware queries, and price-per-token; ChatGPT and Claude still pull ahead on careful writing, technical depth, and multi-step planning.

If you are weighing this against a newer release, the step up is Grok 4.7 at $2 / $6 per million tokens with a 500k context, which scores 46.3 on Intelligence Index v4.3.2, or Grok 4.6 at the same rates and 44.3, which is what the Grok app actually runs. Our Grok 5 leak roundup covers the much larger model still in training. Grok 4.3 stays cheaper at $1.25 / $2.50 with a longer 1M context, so it is still the value pick for long-document work, and the intermediate Grok 4.5 is now deprecated on Artificial Analysis.

Where Grok 4.3 Fits in the Best AI Models Landscape

Across the broader best AI models map, Grok 4.3 carved out a clear niche: cheap frontier-class reasoning with a 1M context and the most price-flexible plan structure on the market at the time. It did not displace GPT-5.5 for raw intelligence or Claude Opus 4.7 for careful analytical work, and two generations of releases on every side have moved that board since. It does, however, make Grok the obvious “second model” to keep on hand for video review, fast file output, and high-volume agentic work where price matters.

For Mac users specifically, the calculus is straightforward. Native Mac apps for ChatGPT, Claude, and Gemini all shipped or got refreshed in April 2026. xAI still ships none. Until xAI ships a Mac client, the cleanest way to run Grok 4.3 next to the others is Fello AI, which bundles Grok, Claude, ChatGPT, Gemini, and DeepSeek into one Mac and iPhone app for $9.99/month. Our Grok desktop client guide for macOS covers the options in full.

Conclusion

Grok 4.3 is the most aggressive pricing play xAI has made yet, and the feature set finally justifies the attention. Real file output and a 1M-token context, still available via the API at $1.25/M input, made it a serious option for daily AI work. You do not need to pay $200 to $300 for ChatGPT Pro or Claude Max to get this kind of capability anymore. It is not the smartest model, but it is the cheapest model that does this much.

Our recommendation: developers should jump straight to the API at $1.25/M input for the cheapest path. Chat users cannot buy Grok 4.3 by name any more: a paid SuperGrok plan buys Expert and Heavy, and both of those modes say Grok 4.6. And if you want Grok next to the other big models on Mac without juggling subscriptions, run them all in Fello AI.

FAQ