MiMo v2.6 Flash
XiaomiBalanced
ThinkingTool UseVisionStructured Output
About this model
MiMo-V2.6-Flash is an open-source foundation model developed by Xiaomi. Built on a Mixture-of-Experts architecture with 309B total parameters and 15B activated per token, it employs a hybrid attention mechanism for...
Performance Tier
Balanced
MiMo v2.6 Flash is a balanced model from Xiaomi : strong performance at a reasonable price.
Strong cost-performance ratio. Reliable for most professional use cases without premium pricing.
Pricing
This model is included in Elosia plansEco
Minimal cost. Ideal for very high volume or simple tasks.
| Type | per 1M tokens |
|---|---|
| Input (prompt) | $0.140 |
| Output (completion) | $0.280 |
| Cache read | $0.0028 |
Capabilities
Context Length1.0M
Max Output Tokens131K
TokenizerOther
Inputtext, image, video, audio
Outputtext
Release DateSeptember 21, 2026
Benchmarks
General Intelligence
MMLU
Not reported
Mathematics
MATH-500
Not reported
Programming
HumanEval
Not reported
Agentic
Terminal-Bench 2.1
87.6%
Where does MiMo v2.6 Flash stand?
Compare its performance index against every other model.
Recommended Use Cases
AnalysisResearchCodingData Extraction
Strengths
- Sparse MoE with 309B total and 15B active parameters, open weights under the permissive MIT license
- Native omni-modal input (text, image, video, audio) over a 1M-token context, sharing the vision and audio encoders of the series flagship
- Same mixed reinforcement-learning recipe as the series flagship, covering coding, general agents, visual tasks and cybersecurity
- Built-in multi-token prediction drafter for faster speculative decoding
Limitations
- The vendor publishes no GPQA Diamond, SWE-bench or MMMU-Pro results, and the only standard score shown is the vendor's own
- Exploit-development and offensive-security tasks are its weakest area in the vendor's own evaluation
Frequently asked questions
Resources
This model may use your data for training
