# Kimi K3.1 Pro (slim)

> Kimi K3.1 Pro is Moonshot AI's open-weight multimodal mixture-of-experts model, post-trained from Kimi K3, with 2.8T parameters, 104B active and a 1M-token context, published in MXFP4 under the modified-MIT Kimi K3.1 Pro Licence.

- Full: https://www.anchorterminal.com/tools/kimi-k3-1-pro.md (~9,450 tokens) · this version ~1,880 tokens · JSON https://www.anchorterminal.com/tools/kimi-k3-1-pro.json · canonical https://www.anchorterminal.com/tools/kimi-k3-1-pro
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-11-10 · pricing https://www.anchorterminal.com/tools/kimi-k3-1-pro#pricing · reviews https://www.anchorterminal.com/tools/kimi-k3-1-pro#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**D · 48.6/100 · rank #130 of 135 open-weight models · #98 in Open-weight models · not agent-ready · confidence low**


Assessment: Moonshot's newest flagship, eight days old when read, for operators with a rack. Only llama.cpp has its model code in a release, with vLLM and SGLang support still on a release candidate and a branch. The licence is permissive below a revenue threshold, and Moonshot hasn't answered a discussion on the repository yet.

## Facts

- Kind: Open-weight model · vendor: Moonshot AI · category: Open-weight models · legal entity: Moonshot AI (the licence names Moonshot AI; the Kimi Platform terms name Moonshot AI PTE. LTD.) · provenance 70/100
- Packages: huggingface `moonshotai/Kimi-K3.1-Pro`
- Auth: None · pricing: Free · x402: no · licence: Kimi K3.1 Pro License (modified MIT)
- Benchmark probes: not run yet, so Performance is pending
- Surface graded: The open weights at huggingface.co/moonshotai/Kimi-K3.1-Pro, read from the model card, the files and their metadata, the licence, vLLM's recipe and SGLang's cookbook for the model, and the runtimes' own code at their latest releases on 10 November 2026. Nothing was downloaded or run. Moonshot's hosted API (platform.kimi.ai) isn't listed in the directory and isn't graded here
- Model: A mixture of experts with 2.8T parameters and 104B active, 93 layers (69 Kimi Delta Attention and 24 gated DeepSeek Sparse Attention), 896 routed experts with 16 per token plus 2 shared, and a 401M-parameter MoonViT-V2 vision encoder. Text and images in. 2,780,110,363,392 parameters in the files per Hugging Face's metadata. The card says it was post-trained from Kimi K3
- Files: 96 safetensors shards, MXFP4 expert weights (stored as U8) with BF16 and F32 for the rest, 1,561.3 GB stored. No pickle files. Hugging Face lists a SHA-256 for every shard. The repository also holds Python for the configuration, tokeniser, processor and prompt encoding, loaded with trust_remote_code
- Licence: Kimi K3.1 Pro Licence, MIT-based, copyright Moonshot AI, for the code and the weights. A model-as-a-service business with more than $20 million in revenue over 12 months needs a separate agreement before commercial use, and products with over 100 million monthly users or $20 million a month in revenue must display "Kimi K3.1-Pro". Internal use and use through Moonshot's own products or certified inference partners are exempt. No acceptable use policy
- Runtimes: llama.cpp b12242 has the kimi-k31 architecture and names this repository in its conversion code. vLLM 0.33.0 and SGLang 0.5.23 carry Kimi K3's model code only. vLLM's recipe needs 0.34.0, a release candidate on 10 November 2026, and SGLang's cookbook a branch or a development image. Transformers 5.22.2 has no kimi_k31 code and loads Moonshot's own from the repository. Ollama has a cloud entry only
- Prompt format: No Jinja chat template in the repository and none in tokenizer_config.json. Prompts are rendered by encoding_k3.py, which Moonshot updated on 5 November 2026
- Tools and reasoning: The card says tool calls and reasoning come back as tool_calls and reasoning_content and that the whole assistant message has to be passed back in later turns. vLLM's recipe and SGLang's cookbook parse both as kimi_k3. reasoning_effort is low, high or max per request, with max the default. The card links guides for structured output, partial mode, tool choice and dynamic tool loading on Moonshot's API
- Context: 1,048,576 tokens, the context length on the card and max_position_embeddings in the text config
- Hardware: The card states no memory needs. vLLM's recipe serves the MXFP4 checkpoint with tensor parallelism 8 on two GB300 trays, one B300 node or one MI355X node. The files are 1,561 GB and don't fit in 512 GB
- Training data: One sentence on the card, large-scale joint reinforcement learning across hundreds of downstream domains on top of Kimi K3. No sources, sizes or knowledge cut-off
- Safety: The card reports Moonshot's own cyber results, 74.7 on ExploitBench and 20.7 on ExploitGym. No independent evaluation of this model was found. The UK AI Security Institute's and CAISI's assessment of July 2026 was of Kimi K3

- Scores: Capability pending, Reach 40, Risk 50, Agent interface 65, Documentation and upkeep 51 · total over the 7 assessed categories
- Why:  · Reach, Memory 0, the only build is Moonshot's 1,561 GB MXFP4 release, far above 256 GB. · Risk, File safety 15, safetensors only, less 10 for remote code and prompts built by the repository's script. · Agent interface, Chat template 10, the format is only code to run. · Documentation and upkeep, Model card 11, no limitations, languages or cut-off.
- Sources: 15, open questions: 8, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/kimi-k3-1-pro.json
- Its vendor verifies the listing with `verify_listing` or `POST https://www.anchorterminal.com/api/v1/verify` (the full twin says how); no effect on the grade.
- Support: Discord https://discord.gg/FBDhVhVH7M; https://huggingface.co/moonshotai/Kimi-K3.1-Pro/discussions; more in the full twin

## Before you call it

1. Use Moonshot's API or Together unless you have a node of eight B300-class accelerators
2. With vLLM use 0.34.0 or later and the recipe's flags, `--tool-call-parser kimi_k3 --reasoning-parser kimi_k3`
3. Pass the whole assistant message back in later turns, reasoning_content and tool_calls included, as the card requires
4. Set `reasoning_effort` to low, high or max per request. Max is the default and thinking can't be turned off
5. Kimi K3's model code doesn't load it, because the attention layers changed
6. Validate JSON output yourself, since the card doesn't document structured output for the weights

## Connect

```bash
# vLLM, per its Kimi-K3.1-Pro recipe linked from the card (vLLM 0.34.0 or later, two GB300 trays)
vllm serve moonshotai/Kimi-K3.1-Pro --trust-remote-code \
  --tensor-parallel-size 8 --nnodes 2 --node-rank 0 --master-addr $HEAD_IP \
  --load-format fastsafetensors --enable-auto-tool-choice \
  --tool-call-parser kimi_k3 --reasoning-parser kimi_k3
```

```bash
curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/Kimi-K3.1-Pro",
    "messages": [{"role": "user", "content": "Hello!"}],
    "reasoning_effort": "high"
  }'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 75.4 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Gemma 4 31B IT | BB | 74.1 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-31b-it.min.md |
| Qwen3-Embedding-4B | BB | 73.6 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 12B IT | BB | 72.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.min.md |
| Gemma 4 26B A4B IT | BB | 72.2 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.min.md |


## Panel reviews (0, desk reviews from public material, no calls made)