Moonshot AI’s new flagship costs a third less than the model it replaces, and on one independent leaderboard it sits right behind Anthropic’s best. Kimi K3.1 Pro went live on Friday, October 23, on Kimi.ai, the Kimi mobile apps, Kimi Work, Kimi Code and the Kimi API at $2 per million input tokens and $10 per million output tokens, the Chinese lab announced on X at 1:15 PM ET. Moonshot calls it its “most capable 3T-class open-weight model.”
There’s a catch in that last phrase. You can’t download Kimi K3.1 Pro yet. Moonshot says the full weights will be released “by Nov 4,” and until then it’s an API model like any other. For developers choosing a model this week, the pricing and the benchmark story are real now. The open part is still a promise with a date on it.
- Price: $2 input / $10 output per million tokens on the Kimi API ($0.20 for cache hits), down from $3 / $15 for Kimi K3. The faster Highspeed version costs $4 / $20.
- Availability: live since October 23 on Kimi.ai, the Kimi apps, Kimi Work, Kimi Code and the API (after a $1 minimum top-up). Enterprise access is off by default until an admin turns it on.
- Open weights: promised “by Nov 4,” per Moonshot. No download and no licence text were published when we checked.
- The catch: its benchmark wins are mostly Moonshot’s own runs. Artificial Analysis independently scores it 55, fourth of 215 models, but finds it slower and wordier than average.
Kimi K3.1 Pro price: a third cheaper than K3
Moonshot’s API platform lists three K3-series models side by side. All of them have a 1,048,576-token context window, according to the pricing docs.
| Model (Kimi API) | Input | Output | Cache hit | Cache write |
|---|---|---|---|---|
| Kimi K3.1 Pro | $2.00 | $10.00 | $0.20 | $2.00 |
| Kimi K3.1 Pro Highspeed | $4.00 | $20.00 | $0.20 | $4.00 |
| Kimi K3 (previous flagship) | $3.00 | $15.00 | $0.30 | $3.00 |
Prices are per million tokens. There’s no surcharge for long prompts: the K3.1 Pro quickstart says pricing is flat across the whole 1M window. The same page says the model always thinks before answering. You can only turn the reasoning effort down (low, high or max, with max the default). Output defaults to 131,072 tokens per call.
That $2 / $10 price matches Google’s Gemini 4 Argon, but only for now. Google’s Gemini API pricing page says Argon’s rate doubles to $4 / $20 on December 1. Moonshot hasn’t called its price a promotion.
What Moonshot says Kimi K3.1 Pro is good at
Moonshot’s tech blog describes K3.1 Pro as a post-trained version of Kimi K3 with “substantially larger post-training compute,” a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active per token. The pitch is coding, 3D work and finance.
On Terminal-Bench 4.0, a test of coding agents in a terminal, Moonshot reports 65.7%, ahead of OpenAI’s GPT-6 Astra at 58.2% and just behind Claude Opus 5.5 at 66.4%. Those rival numbers aren’t Moonshot’s runs: its footnotes say they come from the other labs’ posts or the public leaderboard, while Kimi ran in the Claude Code harness. On BenchCAD, which turns multi-view renders into CAD code, Moonshot puts K3.1 Pro at 92.7% with tools, behind Opus 5.5 (96.2%) and GPT-6 Astra (95.9%).
The finance claim rests on an internal test. On Moonshot’s Kimi Hallu Bench Hard, the claim-level hallucination rate fell from 23.4% for K3 to 8.3%, the source of the “hallucination down by 65%” line in the launch post. Moonshot itself notes the test is built to provoke errors and that the rates “do not represent overall error rates in daily use.”
What an independent leaderboard measured
Artificial Analysis, which runs its own tests, scores Kimi K3.1 Pro at 55 on its Intelligence Index and ranks it fourth of 215 models. In the chart on that page, the only entries above it are Claude Opus 5.5, Claude Sonnet 5.5 and Meta’s Muse Wonder Full. Kimi K3 scored 43.6 on the same chart, so the jump between versions is large.
The same page explains why the sticker price isn’t the whole bill. K3.1 Pro generated 150 million output tokens while running the index, against a median of 87 million, and Artificial Analysis rates it “very verbose.” It measured 73.5 output tokens per second, slower than the 82 average. Moonshot says Highspeed is two to three times faster, but at double the price.
The Kimi K3.1 Pro open weights come later, and the licence isn’t out
Until the weights ship, Artificial Analysis lists K3.1 Pro as a proprietary model. When we checked Moonshot’s Hugging Face page on Saturday morning, the newest repository was still Kimi K3. Moonshot hasn’t said which licence K3.1 Pro will use. If it follows the Kimi K3 licence, use is free, but companies selling model access with more than $20 million in revenue over 12 months need a separate deal, and products with over 100 million monthly users must display the model name.
Running it yourself won’t be casual. At one byte per parameter, 2.8 trillion parameters is roughly 2.8 TB of weights before any memory for context, by our arithmetic. Moonshot hasn’t published hardware requirements.
The weights will also be deliberately weaker at hacking. Moonshot says it removed cyber-related training data “for the open-weight release,” and that on its in-house exploit benchmark the final model solves 81.0% of tasks (best of three tries), versus 90.5% for a checkpoint trained without those cuts. The company says the model stays behind proprietary models that have been public for at least three months. That’s Moonshot’s own test, and nobody outside has checked it yet.
What to watch before November 4
Three dates matter. On October 31, Moonshot’s pricing docs say, the older K2 models stop accepting new users, with all K2 models retired by the end of December. By November 4, the K3.1 Pro weights and their licence should appear. And outside testers will then be able to check Moonshot’s numbers on their own hardware. One small sign the launch was rushed: on Saturday morning, the K3.1 Pro quickstart page still showed unfilled placeholders where its benchmark scores should be.
Frequently asked questions
How much does Kimi K3.1 Pro cost?
On the Kimi API it costs $2 per million input tokens, $10 per million output tokens and $0.20 per million cached input tokens, with no long-context surcharge. Kimi K3.1 Pro Highspeed costs $4 and $20. The previous flagship, Kimi K3, is listed at $3 and $15.
Is Kimi K3.1 Pro open source?
Not yet. Moonshot calls it an open-weight model and says the full weights will be released by November 4, 2026. When we checked on October 24, no download or licence had been published.
What is the Kimi K3.1 Pro context window?
Kimi K3.1 Pro has a 1,048,576-token (1M) context window. Output defaults to 131,072 tokens per request and can be raised to the full window, according to Moonshot’s API docs.
Is Kimi K3.1 Pro better than Claude Opus 5.5?
Not on the numbers published so far. Moonshot’s own chart puts it just behind Opus 5.5 on Terminal-Bench 4.0 (65.7% vs 66.4%), and Artificial Analysis ranks Opus 5.5 above it on its Intelligence Index. It is cheaper per token.
Sources
- Moonshot AI (Kimi), “Introducing Kimi K3.1 Pro” tech blog (October 23, 2026; parameters, benchmarks and footnotes, hallucination test, cyber data removal and xdev-bench-hard results, availability, weights by Nov 4): kimi.ai/blog/kimi-k3-1-pro
- @Kimi_Moonshot on X, K3.1 Pro launch post (October 23, 2026, 1:15 PM ET; 2.8T total / 104B active, “hallucination down by 65%,” open weights by Nov 4, benchmark chart): x.com/Kimi_Moonshot/status/2113680701770068067
- Kimi API Platform home page (K3.1 Pro, Highspeed and K3 prices): platform.kimi.ai
- Kimi API docs, “Model Inference Pricing Explanation” (context window, cache billing, K2 series availability and retirement): platform.kimi.ai/docs/pricing/chat
- Kimi API docs, “Kimi K3.1 Pro” quickstart (always-on thinking, reasoning effort, output limits, flat pricing, $1 top-up): platform.kimi.ai/docs/guide/kimi-k3-1-pro-quickstart
- Artificial Analysis, “Kimi K3.1 Pro: Intelligence, Performance & Price Analysis” (Intelligence Index 55, rank 4 of 215, speed, verbosity, cost per task, chart): artificialanalysis.ai/models/kimi-k3-1-pro
- Google, Gemini Developer API pricing (Gemini 4 Argon $2 / $10 through November 30, 2026, $4 / $20 from December 1): ai.google.dev/gemini-api/docs/pricing
- Hugging Face, moonshotai organization and the Kimi K3 licence: huggingface.co/moonshotai, Kimi K3 LICENSE
- Kimi AI on YouTube, “Meet Kimi K3.1 Pro” (October 23, 2026): youtube.com/watch?v=DIyHTpfWctU
All sources accessed October 24, 2026, between about 8:30 and 9:15 AM ET.
Last updated: October 24, 2026. To report an error, see our corrections page. Articles are drafted with AI assistance and reviewed and edited by an editor; see our editorial policy.
