Moonshot Releases Kimi K3.1 Pro Weights for Self-Hosting
AI · First reported 4 Nov, 15:30 · Updated 16:31
Moonshot AI has published Kimi K3.1 Pro’s weights after first offering the model through its API. The release enables inspection, fine-tuning and self-hosting, but the article says deployment requires substantial hardware and that teams should check the license before use.
- Moonshot says K3.1 Pro keeps the Kimi K3 architecture and uses additional post-training compute, including reinforcement learning across hundreds of domains.
- Moonshot-reported coding benchmarks show mixed comparisons: K3.1 Pro scored 65.7 on Terminal-Bench 4.0 versus Claude Opus 5.5’s 66.4, and 51.7 on SWE-Marathon versus 46.3. The article cautions that harness and tool differences limit model-only conclusions.
- The weights can be self-hosted, but their reported size is about 5.6 TB in BF16 or 2.8 TB at 8-bit precision, before runtime overhead. The article says K3.1 Pro’s separate license had not been announced at publication.
Summary written by Lux AI from 2 articlesUnverified
Key facts
- Weights published
- Moonshot AI published Kimi K3.1 Pro weights on Hugging Face.
- Model size
- Approximately 2.8 trillion total parameters and 104 billion active parameters per token.
- Context window
- 1,048,576 tokens.
- Weight storage estimate
- About 5.6 TB in BF16 or 2.8 TB at 8-bit precision, before runtime overhead and the key-value cache.
- API launch date and prices
- October 23, 2026; $2 per million input tokens and $10 per million output tokens.
- Thinking mode
- The API accepts thinking-mode requests exclusively; disabling thinking returns HTTP 400, according to the article.
Source spectrum
Far left
Left
Lean left
Centre
Lean right
Right
Far right
Unrated
Left 0 · Centre 0 · Right 0 of 0 rated sources · +1 unrated
Background
- An earlier Lux story reported on the cost of running a local large language model, a topic relevant to the infrastructure trade-offs of self-hosting.
- An earlier Lux story covered companies working to control AI agents, another area connected to the article’s focus on coding agents and deployment checks.
All coverage
Coverage over time
LeftCentreRight
Articles per hour since first reported, by side.
Ownership
- Unknown 50% (1)
- Corporation 50% (1)
Share of the 2 sources by owner type.
From the companies
- Kimi K3.1 Pro model weights are out
Hugging Face Blog · 4 Nov, 15:30