Kimi K3.1 Pro, explained

Kimi K3.1 Pro is the reasoning model Moonshot AI announced after Kimi K3, and it went straight into fourth place on the Artificial Analysis Intelligence Index. At launch there was no public API and the weights were not out.

The scores

Artificial Analysis Intelligence Index v4.3.2, checked 24 October 2026. Slingshooter's figure is for the base model, GLM 5.3; uncensored builds are not benchmarked separately.

Only Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full scored higher at launch. Kimi K3.1 Pro gained 11 points on Kimi K3, and most of the gain is agentic: its GDPval-AA score of 1,846 Elo placed it third.

Where the weights stand

The weights were not published at launch. Moonshot has said it will publish them, and Artificial Analysis expects Kimi K3.1 Pro to be the strongest open-weights model when it does. There is a precedent: Kimi K3, the release before it, came out with open weights under its own Kimi K3 License eleven days after it was announced.

Is there an uncensored Kimi K3.1 Pro?

Not yet, from Wild West API or anyone else. Abliteration removes the refusal direction from the weights, and an uncensored finetune retrains them, so both need the weights first. Prompts that talk the official model out of refusing are a different thing, and they get patched. When the weights are out and an uncensored build passes a tool-calling test, it gets added to the line. The uncensored Kimi K3.1 Pro page has the full answer in seventeen languages.

What to use today

The strongest uncensored model you can call is Slingshooter (glm-5.3-outlaw), at 45 on the Intelligence Index. It calls tools, reads a million tokens and answers both the OpenAI and the Anthropic formats.

curl https://wildwestapi.com/v1/chat/completions \
  -H "Authorization: Bearer sk-ww-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3-outlaw",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

FAQ

Related