MiMo v2.5
XiaomiBalanced
ThinkingTool UseVisionStructured Output
About this model
MiMo-V2.5 is a native omnimodal model by Xiaomi. It delivers Pro-level agentic performance at roughly half the inference cost, while surpassing MiMo-V2-Omni in multimodal perception across image and video understanding...
Performance Tier
Balanced
MiMo v2.5 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, audio, image, video
Outputtext
Release DateApril 22, 2026
Benchmarks
General Intelligence
MMLU
Not reported
Mathematics
MATH-500
Not reported
Programming
HumanEval
Not reported
Multimodal
MMMU-Pro
77.9%
Agentic
SWE-bench Pro
56.1%
Terminal-Bench 2.0
65.8%
Recommended Use Cases
AnalysisResearchCodingData Extraction
Strengths
- Native omni-modal input — image, video and audio understanding in one open-weight model, uncommon in its price band
- MoE architecture (310B total, 15B active), open-weight under a permissive MIT license — self-hostable
- Very low pricing at $0.14/M input and $0.28/M output — a fraction of comparable multimodal models
- 1M-token context with native long-context support across modalities
- Strong self-reported document and multimodal understanding (MMMU-Pro 77.9)
Limitations
- Below the MiMo-V2.5-Pro flagship on hard coding and reasoning — positioned as the cheaper multimodal sibling
- Standard text benchmarks published only as chart images; independent per-benchmark verification is still limited
- Smaller ecosystem and tooling maturity than established Western multimodal providers
Resources
This model may use your data for training
