Kimi K3.1 Pro: Moonshot’s Cheaper, Stronger K3, Explained

Moonshot AI released Kimi K3.1 Pro on 23 October 2026 (24 October in China), and it costs less than the model it replaces. The API price is $2 per million input tokens and $10 per million output, down from $3 and $15 for Kimi K3. Kimi K3.1 Pro is also stronger: the independent Artificial Analysis Intelligence Index puts it 4th of every model it tracks, at 55.1, up 11.5 points from K3.

Kimi K3.1 Pro is a post-trained version of K3 with the same 2.8 trillion parameters and a 1-million-token context window, and it ships with a faster Highspeed variant. This guide covers what changed from K3 and where Moonshot's own benchmark numbers disagree with the independent ones. It also covers what K3.1 Pro costs in the API and the Kimi app, how to use it on a Mac today, and why the open weights are only a promise until 4 November.

The Key Takeaways

• Cheaper than K3: $2 / $10 per million tokens, a third below K3's $3 / $15, with the same 2.8T-parameter size.
• 4th on the independent index: Artificial Analysis scores Kimi K3.1 Pro 55.1, behind only Claude Opus 5.5 (57.6), Claude Sonnet 5.5 (56.0) and Muse Wonder Full (55.8).
• Top of Terminal-Bench 4.0: Artificial Analysis measured 67%, the highest score of any model it has tested. Moonshot's own run gave 65.7%.
• Two hallucination numbers: Moonshot reports a drop from 23.4% to 8.3% on its internal stress test. Artificial Analysis measured a smaller fall, from 53% to 44%.
• Weights not out yet: Moonshot promises the full open weights by 4 November 2026. Until then, Kimi K3.1 Pro runs only on Kimi.com, the Kimi apps, Kimi Work, Kimi Code and the Kimi API.

What Is Kimi K3.1 Pro?

Kimi K3.1 Pro is Moonshot AI's newest flagship model and the first model in its K3.1 series. It is a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active per token, a 1-million-token context window, and image input as well as text. Moonshot's API docs also show video files as input. Moonshot calls it "our most capable 3T-class open-weight model to date" in the Kimi K3.1 Pro tech blog.

Kimi K3.1 Pro is not a new architecture. Moonshot describes it as "a post-trained version of Kimi K3, with substantially larger post-training compute", which means the base model is the same one covered in our Kimi K3 explainer. The gains come from reinforcement learning across what Moonshot calls "hundreds of downstream domains", with coding, 3D modelling and finance named as the focus.

Moonshot announced the launch from its official account, listing the size, the context window and the 4 November weights date in one post:

Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.

🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.

Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on http://Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.

Open weights by Nov 4, 2026.

🔗 API: http://platform.kimi.ai
🔗 Tech blog: http://kimi.ai/blog/kimi-k3-1-pro

— Kimi.ai (@Kimi_Moonshot) October 23, 2026

Kimi K3.1 Pro Specs

Kimi K3.1 Pro has the same parameter count and context window as Kimi K3. The table collects its specifications from Moonshot and Artificial Analysis.

Kimi K3.1 vs Kimi K3: What Changed

Kimi K3.1 Pro is the same size as Kimi K3, costs 33% less per token, and scores 11.5 points higher on the Artificial Analysis Intelligence Index. The biggest single jump is on Terminal-Bench 4.0, a test of coding agents working in a real terminal. Artificial Analysis measured Kimi K3 at 13% and K3.1 Pro at 67%.

Kimi K3.1 Pro's lower token price does not make every job a third cheaper. Artificial Analysis found K3.1 Pro uses 14% more output tokens per task than K3, so its cost per Intelligence Index task is $2.11 on Artificial Analysis's model page, against $2.00 for K3. Artificial Analysis's launch post gave $1.98, and the page figure has moved since. You get a stronger model for about the same money per task, not the same model for less.

Kimi K3.1 Benchmarks: Moonshot's Numbers vs Independent Tests

Kimi K3.1 Pro is close to the best proprietary models on coding and agent work, and the independent numbers mostly back Moonshot up. Where the two sets disagree, the gap is in Moonshot's favour on hallucinations and in Artificial Analysis's favour on Terminal-Bench. If benchmark names are new to you, our guide to AI benchmarks explains what each test measures.

Coding and Agent Work

Kimi K3.1 Pro's strongest independent result is agentic work. Artificial Analysis ranks it 3rd on GDPval-AA, a test of real office tasks, at 1,846 Elo. That is up 312 Elo from Kimi K3 and only 20 behind Claude Opus 5.5. On Terminal-Bench 4.0, Artificial Analysis measured 67% and called it "the highest score of any model". Moonshot's own Terminal-Bench run, using the Claude Code harness, gave a slightly lower 65.7%, just under the 66.4% it lists for Claude Opus 5.5. The two runs used different setups, and they disagree on the order: Artificial Analysis puts Kimi K3.1 Pro first, Moonshot's own table puts it 0.7 points behind Claude Opus 5.5. Either way, Kimi K3.1 Pro is level with the best closed models on terminal coding.

Artificial Analysis summed up the launch in one post, including the projection that matters most for open models:

Kimi K3.1 Pro launches at 4th on the Artificial Analysis Intelligence Index scoring 55, making large gains in agentic performance; it will be the leading open-weights model once its weights are published

@Kimi_Moonshot's new reasoning model gains 11 points over Kimi K3 (max, 44). It trails only Claude Opus 5.5 (max with fallback, 58), Claude Sonnet 5.5 (max with fallback, 56), and Muse Wonder Full (max, 56) on the Intelligence Index.

Once the weights for K3.1 Pro are released it will be the highest-scoring open weights model and the gap between the leading open weights and proprietary models on the Intelligence Index will narrow from 12 points to 3.

Key results:
➤ Large gains in agentic performance: Kimi K3.1 Pro scores 3rd on GDPval-AA with 1,846 Elo, up 312 Elo from Kimi K3, behind only Claude Opus 5.5 (1,866 Elo) and Muse Wonder Full (1,857 Elo). Kimi K3.1 Pro also achieves the highest score of any model on Terminal-Bench 4.0. K3.1 Pro scores 67%, up from 13% for Kimi K3 and ahead of Claude Sonnet 5.5 (64%). AutomationBench-AA, which tests agentic SaaS workflows, rises from 58% to 73%.

➤ Gains in other evals are more moderate: Humanity's Last Exam rises from 47% to 52%, CritPt from 23% to 25%, and SciCode from 59% to 61%. AA-Omniscience Index rises from 20 to 30, as the hallucination rate falls from 53% to 44% and accuracy rises from 48% to 51%.

➤ Cost per Task roughly equal to Kimi K3 at $1.98 per Task: K3.1 Pro uses 55K output tokens per Intelligence Index task, 14% more than Kimi K3 at 48K, while its first-party token prices are 33% lower.

Additional model details:
➤ Context window: 1M tokens.
➤ Size: 2.8T total, 104B active parameters
➤ Input modalities: Text and image
➤ Weights and license: Weights currently not published, Moonshot is expected to release the weights soon after launch
➤ Pricing: $2/$10/$0.20 per 1M input/output/cache hit tokens, 33% lower than Kimi K3 on all three rates.

Congratulations to the team at @Kimi_Moonshot on the release!

— Artificial Analysis (@ArtificialAnlys) October 23, 2026

Hallucinations: Two Very Different Numbers

Kimi K3.1 Pro makes fewer factual errors than K3, but how many fewer depends on who is measuring. Moonshot's internal Kimi Hallu Bench Hard shows the claim-level hallucination rate falling from 23.4% to 8.3%, which is the "hallucination down by 65%" in the launch post. Moonshot's own API quickstart gives a different result, 23.35% to 6.85%, and describes a different test set built from conversations where users had flagged factual errors. Even the vendor's figure is not settled. Moonshot adds its own caveat. The test is built from "examples deliberately crafted to probe situations where models are prone to errors", and "the reported rates do not represent overall error rates in daily use."

The independent AA-Omniscience test shows a real but smaller improvement. Artificial Analysis measured Kimi K3.1 Pro's hallucination rate at 44%, down from 53%, with accuracy up from 48% to 51%. The two tests measure different things, so neither number is wrong. The practical reading: Kimi K3.1 Pro is more careful than K3, and it still needs checking on facts that matter.

Where Kimi K3.1 Pro Still Trails

Kimi K3.1 Pro still trails Claude Opus 5.5 overall, by 2.5 points on the Intelligence Index. Moonshot's own BenchCAD table also puts it behind both Claude Opus 5.5 and GPT-6 Astra on 3D reconstruction. Artificial Analysis also measured the gains outside agent work as "more moderate": Humanity's Last Exam rose from 47% to 52% and SciCode from 59% to 61%. For a closer look at how the Kimi line compares with Anthropic's models, see our Kimi vs Claude comparison.

Kimi K3.1 Pro Pricing

Kimi K3.1 Pro costs $2.00 per million input tokens and $10.00 per million output tokens on the Kimi API, with cached input at $0.20. The Highspeed variant costs twice as much. Both prices come from Moonshot's Kimi API pricing page, checked on 23 October 2026.

API Pricing

Kimi K3.1 Pro has flat pricing across the full 1-million-token context, with no surcharge for long prompts. The Kimi API also turns on context caching automatically for repeated prompt prefixes. Moonshot treats K3.1 Pro as a flagship model, so the API unlocks only after a top-up of at least $1. Zero Data Retention is available for API customers.

Kimi App Plans

In the Kimi app, the faster Kimi K3.1 Pro Highspeed is listed on the Pro plan and above. The Kimi pricing guide covers each plan in detail. On Moonshot's membership page on 23 October 2026, monthly prices were Plus $19, Pro $39, Max $99 and Ultra $199, or $15, $31, $79 and $159 a month billed annually. The Plus plan does not list Highspeed access. Moonshot's page does not say whether the free Adagio plan gets the standard Kimi K3.1 Pro, so check the model picker in the app before you pay.

How to Use Kimi K3.1 Pro on a Mac

You can use Kimi K3.1 Pro on a Mac today in four ways: in a browser, in the Kimi Work desktop app, in Kimi Code, or through the API. All four are listed in Moonshot's availability notes.

1. In the browser: open kimi.com and sign in. Kimi K3.1 Pro rolls out on Kimi.com and the Kimi mobile apps first.
2. Kimi Work: download the Kimi Work desktop app, version 3.1.0 or later. Moonshot lists it for Windows and Apple silicon Macs, so an Intel Mac needs the browser instead.
3. Kimi Code: run Kimi Code in Terminal and type /model to switch to Kimi K3.1 Pro.
4. The API: top up at least $1 on the Kimi API platform and call kimi-k3.1-pro or kimi-k3.1-pro-highspeed.

Kimi Enterprise works differently. Administrators must switch Kimi K3.1 Pro on for their organisation, because access is off by default at launch.

Kimi Without Another Account

Trying a new model usually means another account, another app and another bill. Fello AI skips that step. Kimi sits in the same native Mac and iPhone app as ChatGPT, Claude, Gemini, DeepSeek and Grok, and new models usually reach Fello AI as soon as their API opens. You can ask Kimi and Claude the same question in one window and keep whichever answer is better.

Kimi K3.1 Open Weights: Promised for 4 November

The Kimi K3.1 Pro weights are not public yet. Moonshot says "the full model weights will be released by Nov 4", and Artificial Analysis currently lists the model as proprietary with "weights currently not published". Until the weights land on Hugging Face, Kimi K3.1 Pro is a closed model you can only use through Moonshot's own apps and API.

The date matters because of Artificial Analysis's projection. Once the weights ship, Kimi K3.1 Pro becomes the highest-scoring open-weights model on its index. The gap between the best open model and the best proprietary one would shrink from 12 points to 3. Moonshot has kept this kind of promise before. It launched Kimi K3 on 16 July, promised the weights by 27 July, and published them on Hugging Face on 27 July. Our ranking of the best Chinese AI models will move when the K3.1 Pro weights are confirmed.

Moonshot Removed Hacking Data Before the Open Release

Moonshot removed cyber-related training data from Kimi K3.1 Pro before the open-weight release, and published the effect. On its in-house xdev-bench-hard test of exploit development, a checkpoint trained without the removal solved 90.5% of tasks at pass@3. The final Kimi K3.1 Pro solves 81.0%. Moonshot says the model's offensive cyber capability "remains behind that of frontier proprietary models that have been publicly available for at least three months". It also says it aims "not to release open weights that would extend the absolute frontier of cyber capability." Both figures come from Moonshot's own test, and no outside lab has checked them yet.

The Verdict

Kimi K3.1 Pro is the model to use if you already run Kimi K3. It costs a third less per token, sits 2.5 points behind Claude Opus 5.5 on the independent index, and tops Artificial Analysis's Terminal-Bench 4.0 run. On the API, moving from K3 means changing the model ID to kimi-k3.1-pro, and the cost per task stays about the same. Test it on your own prompts before you switch a production workload. Treat Moonshot's 8.3% hallucination figure as a best case, keep checking facts that matter, and watch 4 November. If the weights ship on time, Kimi K3.1 Pro becomes the strongest open model anyone can download.

FAQ