Introducing Kimi K3.1 Pro

What it claims

Kimi K3.1 Pro is a Moonshot AI model served on the Kimi API since 23 October 2026. Simon Willison notes it is the first (soon-to-be) open weight model to score this close to the frontier. Moonshot states "Full model weights will be released by Nov 4, 2026." At $2.00 per million input tokens and $10 per million output tokens, it costs the same as GPT-6.1 Sol and half as much as Claude Opus 5.5. The Highspeed variant runs 2-3x faster at twice the price. Input token cache hits are the exception: they cost the same on both variants.

Evidence

Artificial Analysis scores Kimi K3.1 Pro at 55. Claude Opus 5.5 is the only model ahead of it, at 58. Claude Fable 5.1, GPT-6 Astra and Gemini 4 Argon all score 53. Moonshot's own benchmarks put the model at about the same level as Opus 5.5 and GPT-6 Astra. By those numbers it beats both on a few benchmarks and comes close on others. The post does not name those benchmarks or give their figures. Willison's own testing is qualitative: pelican-on-a-bicycle SVGs at each reasoning level, plus fresh pelicans from the earlier Kimi K3 for comparison.

How to use it

The API is OpenAI-compatible at https://api.moonshot.ai/v1, with the model ID kimi-k3.1-pro. It supports the Responses API and three reasoning effort levels: low, medium and max. Willison ran it with the llm openai endpoint command, which was added in LLM 0.32 in August 2026:

llm openai endpoint \
  https://api.moonshot.ai/v1 \
  --key kimi \
  -m kimi-k3.1-pro \
  "Generate an SVG of a pelican riding a bicycle" \
  --responses \
  -o reasoning_effort low \
  > kimi-k3.1-pro-pelican-low.stdout.txt \
  2> kimi-k3.1-pro-pelican-low.stderr.txt

The command does not log results, so the redirects save the output (stdout) and the reasoning trace (stderr).

Limits

The weights are not out yet, and the post gives no licence, parameter count or context length. The benchmark claims come from Moonshot itself and from a single Artificial Analysis score. The post includes no latency figures for the Highspeed variant beyond Moonshot's "2-3x faster".