Kimi K3.1 Pro: a Chinese model nearly at the frontier — and open weights are coming

Moonshot has launched Kimi K3.1 Pro, now available through its API. For the moment it is a closed model, but the company promises to publish the full weights by November 4, 2026. If that promise holds, Kimi would become the first open-weight model to come this close to the industry's frontier.

Pricing and options

The model costs $2 per million input tokens and $10 per million output tokens — the same as GPT-6.1 Sol and half the price of Claude Opus 5.5. There is also a "Highspeed" variant that runs 2–3x faster but costs twice as much; the one exception is input token cache hits, which are priced the same.

Benchmark results

Moonshot's own benchmarks place Kimi roughly level with Opus 5.5 and GPT-6 Astra — beating them on some tests, coming close on others. More telling is the independent Artificial Analysis score of 55. That puts Kimi ahead of Claude Fable 5.1, GPT-6 Astra and Gemini 4 Argon (all at 53), trailing only Claude Opus 5.5 at 58.

As the author notes, this is the first time a (soon-to-be) open-weight model has scored this highly against the frontier.

The pelican test

Simon Willison tried the model via the Moonshot API using the llm openai endpoint command he added to LLM 0.32 back in August. The command can send prompts to any OpenAI-compatible endpoint, but it doesn't log the results — hence the need to redirect the output and the reasoning trace into files by hand.

The prompt was his signature one: generate an SVG of a pelican riding a bicycle. The model supports three reasoning levels — low, medium and max — and for comparison he ran the same test against the earlier Kimi K3.

Source: Simon Willison