← The Polylog AI Intelligence Brief
Morning Edition · Friday, November 6, 2026Published at 5:11 AM EST · New York
Moonshot AI publishes the weights for Kimi K3.1 Pro, a 2.8-trillion-parameter mixture-of-experts model
The model activates 104 billion parameters per token and ships in a 4-bit format from quantization-aware training, and an independent index scores it 55, three points behind Claude Opus 5.5.
Moonshot AI, the Beijing lab behind the Kimi models, published the full weights for Kimi K3.1 Pro on Hugging Face. Most reports date the upload to November 4, which was the deadline Moonshot set when it launched the model through its API on October 23. The repository is public with no access form or waitlist, according to secondary coverage.
The model is a mixture-of-experts design with 2.8 trillion total parameters, 104 billion active per token, and a context window of about one million tokens. The weights use a 4-bit microscaling floating-point format (MXFP4) with 8-bit activations, produced by quantization-aware training. The download is about 1.56 TB across 96 safetensors shards. One report estimates that serving it takes about eight NVIDIA B300 GPUs, so in practice the weights are open to operators with data-center hardware.
Moonshot's own benchmark table, as compiled by MindStudio, puts it at 70% on Humanity's Last Exam (HLE) with tools, against 67.7% for Claude Opus 5.5 and 57.2% for GPT-6 Astra. On Terminal-Bench 4.0 it scores 65.7 against Opus 5.5's 66.4. It trails substantially on ProgramBench, at 31.5 against 65.0. Moonshot ran its agentic coding rows inside the Claude Code harness. The more neutral number comes from Artificial Analysis, which scores the model 55, an 11-point gain over Kimi K3 and fourth overall behind Opus 5.5 at 58.
The license is Moonshot's own, not a standard open-source license. Reports say it is permissive for most users, but hosts earning more than $20 million a year need a separate agreement. API pricing is $2 per million input tokens and $10 per million output tokens.
NVIDIA has its own reasons to promote open weights. Its blog this week profiled Chris Alexiuk, a Toronto-based product research engineer who says open-source AI should be promoted "from the rooftops." Every large open checkpoint that enterprises host themselves increases demand for GPUs that enterprises buy directly, whoever trained the model.
Veracity: Plausible
72/100Part of a tracked trend
Chinese Open-Weight Models Emerge as the Non-US AI Stack
As Washington restricts foreign access to US frontier models, governments and enterprises cut off from American AI increasingly standardize on downloadable Chinese open-weight models, splitting the world into competing AI supply blocs rather than a single frontier.
Townsquare
Start a discussion in Townsquare.
More from this edition
- OpenAI rates its new GPT-6 Cyber model "Critical" for cyber risk and limits it to vetted security teams
- Anthropic moves toward a November roadshow for its Nasdaq listing with no price range filed yet
- OpenAI pledges to pay grid costs for its own data-center campuses as NVIDIA sells Vera CPUs on agents per watt
- OpenAI reports an unreleased model scoring 0.875 on BenchCAD 2.0, against 0.594 for GPT-6 Astra
- Octopus paper shares one KV cache across sub-agents and lifts BrowseComp-Plus accuracy to 49.6% on a 4-billion-parameter model
- Google cuts Gemini Omni video pricing 30% to $0.07 per second and extends clips to 15 seconds
- Claude Code sent users' account email addresses to websites, and Anthropic answered with an opt-out privacy module
Comments
0No comments yet.