The official Kimi account @Kimi_Moonshot announced the Kimi K3.1 Pro model weights at 16:00 UTC on November 4; the moonshotai/Kimi-K3.1-Pro repository is public and not gated. The model card describes a natively multimodal MoE with 2.8T total and 104B active parameters and a 1M-token context, shipped as MXFP4 weights from quantization-aware training, about 1.56TB in total; vLLM, SGLang and TokenSpeed are the recommended engines. The license allows free use and modification, but Model-as-a-Service businesses with over $20M in annual revenue need a separate agreement.
Key Takeaways
- ✓Architecture: 93 layers (1 dense), 69 KDA plus 24 Gated DSA attention layers, 896 experts with 16 selected per token plus 2 shared experts, and a 401M MoonViT-V2 vision encoder.
- ✓Compared with K3, the full-attention Gated MLA layers are replaced with Gated DeepSeek Sparse Attention for more efficient long-context modeling.
- ✓Vendor-reported scores on the model card: Terminal-Bench 4.0 65.7 (Opus 5.5 66.4), SWE Marathon v1.1 51.7 (Opus 5.5 46.3), HLE-Full with tools 70.0 and BrowseComp 92.1; ProgramBench 31.5 trails Opus 5.5 at 65.0.
- ✓License: a Model-as-a-Service business with over $20M revenue in any 12 months needs a separate agreement with Moonshot, and products above 100M monthly users or $20M monthly revenue must prominently display "Kimi K3.1-Pro"; internal use is exempt from both.
Key Decision Metrics at a Glance
Not enough VRAM? Compare cloud API and self-hosting costs
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and industry pain points
When Kimi K3.1 Pro launched in late October it was available only through the API and Kimi's own products. At 16:00 UTC on November 4 the official Kimi account announced the full weights, and the Hugging Face repository moonshotai/Kimi-K3.1-Pro is now public. For teams that want to self-host a top open model or keep data on their own hardware, it is one of the largest open-weight models available.
Architecture highlights and mechanisms
Per the model card, the MoE has 2.8T total and 104B active parameters across 93 layers, 69 of them KDA and 24 Gated DSA, with 896 experts and 16 selected per token, a 1M-token context, and text plus image input. Compared with K3, the full-attention Gated MLA layers are replaced with Gated DeepSeek Sparse Attention. Weights are MXFP4 with MXFP8 activations from quantization-aware training, about 1.56TB in the repository.
Benchmarks and measured comparisons
All figures are Moonshot's own model-card numbers: Terminal-Bench 4.0 65.7 (GPT-6 Astra 58.2, Claude Opus 5.5 66.4), SWE Marathon v1.1 51.7, DeepSWE v1.1 75.5, HLE-Full with tools 70.0 and BrowseComp 92.1. ProgramBench at 31.5 is well below Opus 5.5 at 65.0. No independent third-party reruns have been published yet.
Developer adoption and getting started
Moonshot recommends vLLM, SGLang or TokenSpeed, each with a K3.1 Pro recipe, or the OpenAI/Anthropic-compatible API on platform.kimi.ai as kimi-k3.1-pro. Multi-turn and tool-calling requests must pass back reasoning_content and tool_calls unchanged. The license allows modification and commercial use, but Model-as-a-Service businesses with over $20M annual revenue need a separate agreement and very large products must display the model name.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.