MiMo v2 Pro
XiaomiFlagship
ThinkingTool Use
About this model
MiMo-V2-Pro is Xiaomi's flagship foundation model, featuring over 1T total parameters and a 1M context length, deeply optimized for agentic scenarios. It is highly adaptable to general agent frameworks like...
Performance Tier
Flagship
MiMo v2 Pro is a flagship model from Xiaomi : the most capable in their lineup.
Best-in-class model from this provider. Highest performance across benchmarks, ideal for demanding tasks.
Pricing
This model is included in Elosia plansModerate
Moderate cost. A balanced choice for regular use without constant cap watching.
| Type | per 1M tokens |
|---|---|
| Input (prompt) | $1.00 |
| Output (completion) | $3.00 |
| Cache read | $0.200 |
Capabilities
Context Length1.0M
Max Output Tokens131K
TokenizerOther
Inputtext
Outputtext
Release DateMarch 18, 2026
Benchmarks
General Intelligence
MMLU
Not reported
GPQA Diamond
83.7%
Mathematics
MATH-500
Not reported
AIME 2025
94.1%
Programming
HumanEval
Not reported
SWE-bench Verified
73.4%
Reasoning
IFEval
Not reported
Agentic
Terminal-Bench 2.0
86.7%
Recommended Use Cases
CodingAnalysisResearchMathematics
Strengths
- Industry-leading Terminal-Bench 2.0 (86.7) — far ahead of Opus 4.6 (65.4) and GPT-5.2 Codex (77.3)
- Exceptional math reasoning (AIME 2025: 94.1%) at a fraction of frontier model cost
- 1M token context with hybrid SWA attention — ~6x KV-cache reduction for efficient long-doc processing
- MoE architecture (1T+ params, 42B active) delivering frontier performance at $1/$3 per M tokens
Limitations
- Text-only — no multimodal support (images/audio/video handled by separate MiMo-V2-Omni)
- Relatively high hallucination rate (30%) compared to top Western models
- No standard academic benchmarks published (MMLU, HumanEval, MATH-500) — opaque on general knowledge
- Proprietary weights, API-only — limited ecosystem and community support
Resources
This model may use your data for training
