Moonshot's own benchmarks show Kimi K3.1 Pro approximately level with Claude Opus 5.5 and GPT-6 Astra, edging ahead on some tests and near on others. Independent analysis by Artificial Analysis gave the model a score of 55, placing it ahead of Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon (each at 53) and behind only Claude Opus 5.5 at 58. This makes Kimi K3.1 Pro the most capable model that is expected to become open-weight. The company also offers a faster 'Highspeed' variant at twice the standard price, except for input token cache hits, which are priced the same. Simon Willison tested the model's reasoning by generating SVG images of a pelican riding a bicycle at three reasoning levels (low, medium and max) and found it responsive across all levels. The open-weight release, if delivered, would be a notable step for AI accessibility, arriving amid broader competition in the field.
AIDeveloping
Moonshot AI Releases Kimi K3.1 Pro on API, Pledges Open Weights by November
New model competes with top AI rivals at half the price
Moonshot AI has placed its latest large language model, Kimi K3.1 Pro, on its API, with a promise to release the full model weights by November 4, 2026. The model is priced at $2 per million input tokens and $10 per million output tokens, matching GPT-6.1 Sol and half the price of Claude Opus 5.5, according to a report by Simon Willison.
score
Kimi K3.1 Pro is the first soon-to-be-open-weight model to score this high. Source: Artificial Analysis
§
Analysis
Why This Matters
- Kimi K3.1 Pro is the first (soon-to-be) open-weight model to match the performance of top proprietary models, potentially broadening access to frontier AI.
- Its competitive pricing could pressure other providers to lower costs.
- The release showcases growing AI capabilities from Chinese labs.
Background
Moonshot AI is a Chinese artificial intelligence company known for its Kimi series of large language models. The K3 series has previously been released with open weights, and K3.1 Pro represents a major performance leap.
What to Watch
- Whether Moonshot meets its Nov 4 deadline for open weights release and the practical usability of the model.
- Subsequent independent benchmarks confirming or challenging the model's performance claims.
- Reactions from other AI companies, especially regarding pricing and openness.
Sources
- Introducing Kimi K3.1 Pro · Simon Willison's Weblog
Checked
10 claims in this story checked against 1 source: 6 supported.
Show each claim
- SupportedMoonshot AI released Kimi K3.1 Pro on its API.Simon Willison's Weblog: “Kimi K3.1 Pro landed on the Kimi API today”
- SupportedMoonshot promised to release the full model weights by November 4, 2026.Simon Willison's Weblog: “Moonshot promise "Full model weights will be released by Nov 4, 2026".”
- Partly supportedKimi K3.1 Pro competes with top AI rivals at half the price.Simon Willison's Weblog: “The US dollar pricing is $2.00/million tokens for input and $10/million for output - the same as GPT-6.1 Sol and half the price of Claude Opus 5.5.”
- SupportedKimi costs $2 per million input tokens and $10 per million output tokens, matching GPT-6.1 Sol and costing half as much as Claude Opus 5.5.Simon Willison's Weblog: “The US dollar pricing is $2.00/million tokens for input and $10/million for output - the same as GPT-6.1 Sol and half the price of Claude Opus 5.5.”
- SupportedMoonshot's benchmarks show Kimi approximately level with Claude Opus 5.5 and GPT-6 Astra, beating them on some tests and coming close on others.Simon Willison's Weblog: “Moonshot's own benchmarks show Kimi as approximately level with Opus 5.5 and GPT-6 Astra - beating them on a few benchmarks, coming close on others.”
- SupportedArtificial Analysis scored Kimi 55, ahead of Claude Fable 5.1, GPT-6 Astra and Gemini 4 Argon at 53 each, and behind only Claude Opus 5.5 at 58.Simon Willison's Weblog: “Artificial Analysis scored Kimi at 55 - ahead of Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon (all at 53) and behind only Claude Opus 5.5 (58).”
- Not in the sourcesKimi K3.1 Pro is the most capable model expected to become open-weight.Simon Willison's Weblog
- SupportedThe Highspeed variant is faster and costs twice the standard price, except input-token cache hits cost the same.Simon Willison's Weblog: “They’re also offering a "Highspeed" version that’s 2-3x faster, at twice the price (except for input token cache hits, which are priced the same).”
- Not in the sourcesSimon Willison found the model responsive when generating pelican-on-a-bicycle SVGs at low, medium and max reasoning levels.Simon Willison's Weblog: “Here are the pelicans I got for Kimi K3.1 Pro on the three reasoning levels it supports - low, medium, and max.”
- Not in the sourcesThe open-weight release would be a notable step for AI accessibility and is arriving amid broader competition in the field.