Skip to content
Main Site News Console

Kimi K3.1 Pro Officially Released

· Simon Willison Translated
技巧LLM

October 23, 2026 - Linklog

Kimi K3.1 Pro is out (source) Kimi K3.1 Pro is available on the Kimi API today—not yet with open weights, but Moonshot promises that “the full model weights will be released by November 4th, 2026.”

They are also offering a “Highspeed” version that runs 2–3x faster at twice the price (except for cached input tokens, which cost the same).

Its USD pricing is $2.00 per million input tokens and $10 per million output tokens—matching GPT-6.1 Sol and coming in at half the price of Claude Opus 5.5.

Moonshot’s own benchmarks show Kimi roughly tied with Opus 5.5 and GPT-6 Astra—edging ahead on a few benchmarks and remaining competitive across the rest.

Artificial Analysis scores Kimi at 55—ahead of Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon (all at 53), trailing only Claude Opus 5.5 (at 58).

This is the first time a (soon-to-be) open weights model has scored this high against top frontier models.

I took it for a spin via the Moonshot API using the llm openai endpoint command, which was added to LLM 0.32 back in August:

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

llm openai endpoint lets you run prompts against any OpenAI-compatible endpoint, but it does not log results by default, which is why the extra arguments are needed to capture stdout and stderr (the reasoning trace).

Here are the pelicans generated by Kimi K3.1 Pro across its three supported reasoning levels (low, medium, and max). Here is max:

Figure 1: Pretty solid effort—feet on the pedals, a well-formed bicycle frame, and a convincing beak. Both legs are on the same side of the frame, though.

For comparison, I also generated fresh pelicans for Kimi K3 across those same three reasoning levels.