# MiMo-V2.6-Flash-RL (slim)

> MiMo-V2.6-Flash-RL is Xiaomi's open-weight agent model, a mixture of experts of 309B parameters with 15B active, taking text, images, video and audio with a 1M context. MIT. The weights are 178 GB at 4 bits, so it needs a server.

- Full: https://www.anchorterminal.com/tools/mimo-v2-6-flash-rl.md (~10,150 tokens) · this version ~1,680 tokens · JSON https://www.anchorterminal.com/tools/mimo-v2-6-flash-rl.json · canonical https://www.anchorterminal.com/tools/mimo-v2-6-flash-rl
- 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/mimo-v2-6-flash-rl#pricing · reviews https://www.anchorterminal.com/tools/mimo-v2-6-flash-rl#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 61/100 · rank #75 of 160 open-weight models · #59 in Open-weight models · not agent-ready · confidence low**


Assessment: A large agent model with an open licence: text, images, video and audio, a 1M context, tool calling it was trained for, and code in four released runtimes. It's a server model, 178 GB at 4 bits, and the card's commands run the repository's code. No safety results, little said about data, and few questions answered.

## Facts

- Kind: Open-weight model · vendor: Xiaomi (MiMo team) · category: Open-weight models · legal entity: Xiaomi (the card and report name Xiaomi MiMo; no LICENSE file names a copyright holder) · provenance 58/100
- Packages: huggingface `XiaomiMiMo/MiMo-V2.6-Flash-RL`
- Auth: None · pricing: Free · x402: no · licence: MIT (metadata only, no LICENSE file)
- Benchmark probes: not run yet, so Performance is pending
- Surface graded: The open weights at huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL, read from the model card, the files and their metadata, the config, the chat template, the commit history, ggml-org's GGUF repository and the runtimes' own code at their latest releases on 10 November 2026. Xiaomi's company facts come from our MiMo-V2.6 Distill listing. The technical report in the repository, the repository's Python files, and the SGLang cookbook and vLLM recipe the card links weren't read. Nothing was downloaded or run
- Model: 310,756,322,688 parameters in the files. The card says a sparse mixture of experts with 309B in total and 15B active, a 681M vision encoder, audio encoders of 308M and 127M, and a five-layer decoder for multi-token prediction. The backbone interleaves sliding-window and global attention
- Files: 67 safetensors files (177.74 GB), with 302.8B parameters stored as packed 4-bit values and the rest in BF16 and FP8, chat_template.jinja, config and tokenizer files, an audio tokenizer, three Python modules, the technical report as a PDF, and a dflash folder with a draft model whose mask_embedding.pt is a pickle. Hugging Face lists a SHA-256 for the large files
- Licence: MIT in the card's metadata. There is no LICENSE file. The card's Use Restrictions section lists prohibited uses and says it is an informational reminder that doesn't amend the licence or add conditions
- Runtimes: The card documents SGLang, pointing to SGLang's MiMo cookbook, and vLLM, pointing to vLLM's recipe and a pre-built image, both with trust-remote-code. Released SGLang 0.5.23 and vLLM 0.33.0 have the model and the mimo parsers, Transformers 5.22.2 has a mimo_v2_flash model, and llama.cpp b12242 has the architecture, with ggml-org's GGUF
- Context: 1M tokens per the card, and 1,048,576 positions in config.json
- Tools and reasoning: The card's SGLang and vLLM commands name the mimo tool-call parser and the mimo reasoning parser, with auto tool choice on vLLM. The chat template takes tools and an enable_thinking switch and keeps reasoning_content. The card itself shows no request with tools
- Hardware: No memory figure on the card. Its SGLang command uses 8 GPUs with tensor parallelism and its vLLM command 4. The files are 177.7 GB
- Training data: The card describes the training method at length (mixed reinforcement learning with graders, then distillation from teachers) and names no data sources, sizes or cut-off. The technical report wasn't read

- Scores: Capability pending, Reach 65, Risk 45, Agent interface 85, Documentation and upkeep 59 · total over the 7 assessed categories
- Why:  · Reach, Memory 5, the smallest build at 4 bits or better is ggml-org's MXFP4 GGUF, 168.93 GB with its projector, and 168.93 GB × 1.2 = 202.7 GB, wit… · Risk, File safety 5, the weights are safetensors, but the dflash folder's mask_embedding.pt is a pickle beside them (15). · Agent interface, Chat template 20, chat_template.jinja in the repository (3,868 characters), with a copy in tokenizer_config.json. · Documentation and upkeep, Model card 14, Intended use, the architecture, the parameters, the context and the languages (in the metadata) are on the card.
- Sources: 13, open questions: 6, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/mimo-v2-6-flash-rl.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: more in the full twin

## Before you call it

1. Plan for a provider unless you have a multi-GPU server: Xiaomi's API platform, OpenRouter or Novita
2. On SGLang or vLLM pass `--tool-call-parser mimo --reasoning-parser mimo`
3. Use temperature 1.0 and top_p 0.95, as the card recommends
4. The card's commands pass trust-remote-code. Read the repository's three Python files first, or use Transformers 5.22.2, which has the model built in
5. Skip the dflash folder unless you need speculative decoding. Its mask_embedding.pt is a pickle
6. The Use Restrictions section asks for human oversight of high-risk actions

## Connect

```bash
pip install "sglang>=0.5.23"
sglang serve --trust-remote-code --model-path XiaomiMiMo/MiMo-V2.6-Flash-RL --tp 8 --reasoning-parser mimo --tool-call-parser mimo
# the card's full command adds data-parallel attention and speculative decoding flags
```

```bash
# SGLang and vLLM serve an OpenAI-compatible API. The card recommends temperature 1.0 and top_p 0.95,
# and names the mimo tool-call and reasoning parsers for both
```

## 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)