XiaomiMiMo/MiMo-V2-Flash resource

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1368 stars · 64 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

MiMo-V2-Flash is a 309B-parameter Mixture-of-Experts language model (15B active) from Xiaomi designed for efficient reasoning, coding, and agentic workflows. It uses a hybrid sliding-window/global attention architecture and Multi-Token Prediction to cut inference costs while supporting 256k context.

Use cases

• run an efficient open-weight LLM for reasoning tasks
• serve a MoE model with low inference cost
• build coding agents that score well on SWE-Bench
• accelerate LLM rollout generation for RL training
• process long documents with 256k context
• self-host a fast code assistant model

When to choose

• you need an open Apache-2.0 MoE model balancing reasoning, coding, and agentic performance
• inference cost and output speed matter more than peak quality
• you need very long context with reduced KV-cache memory

When to avoid

• you need a small model that fits on consumer hardware
• you want a mature ecosystem with broad tooling support
• you need multimodal (image/audio) capabilities

Facets

learning-resource · maturity active

llm-inference machine-learning llm-training agent-framework large-language-models artificial-intelligence deep-learning python cloud mixture-of-experts reasoning-model multi-token-prediction sliding-window-attention code-generation agentic-rl xiaomi ai-agents gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="XiaomiMiMo/MiMo-V2-Flash")