Xiaomi's MiMo V2.5 Pro, released April 22, 2026, has scored 54 on the Artificial Analysis Intelligence Index, tying for first place among open-weights AI
Model details
Xiaomi's MiMo V2.5 Pro, released April 22, 2026, has scored 54 on the Artificial Analysis Intelligence Index, tying for first place among open-weights AI
Two open-source Chinese AI models just outbenchmarked Claude Opus 4.6 and the Western AI industry should be paying close attention
Xiaomi Releases MiMo-V2.5-Pro and MiMo-V2.5: Matching Frontier Model Benchmarks at Significantly Lower Token Cost
MiMo-V2.5-Pro by Xiaomi. 1.0M context. from $1/M tokens. released Apr 2026. See real outputs and compare on Rival.
See performance metrics across providers for Xiaomi: MiMo-V2.5-Pro - MiMo-V2.5-Pro is Xiaomi’s flagship model, delivering strong performance in general agentic capabilities, complex software engineering, and long-horizon tasks, with top rankings on benchmarks such as ClawEval, GDPVal, and SWE-bench Pro. It can independ
This exact model name is also listed by 6 other providers.
Keep Reviews Moving
When AI speeds up shipping, review queues get exposed fast. CodeRabbit reviews pull requests quickly, catches issues that surface late, and adds coverage before code reaches production.
Developers already feel this
MiMo-V2.5-Pro is a large-scale mixture-of-experts model built to handle the most demanding agentic and software engineering challenges. It features a hybrid attention architecture that interleaves sliding window and global attention in a the listed price:the listed price ratio, which significantly reduces memory overhead while maintaining performance over its massive context window. By utilizing learnable attention sink bias, the model ensures logical consistency and strong instruction following even during extended sessions that require thousands of sequential tool calls.
The model incorporates three lightweight multi-token prediction modules using dense feed-forward networks, a design choice that triples output speed and supports efficient rollout during reinforcement learning training. With 1.02 trillion total parameters and 42 billion active parameters, it is optimized for high-performance deployment on specialized silicon. This architecture allows the model to excel in agentic frameworks, making it a robust choice for developers building systems that require autonomous, long-horizon task completion and deep data analysis.
Why teams adopt it
Discuss this model
Add corrections, implementation notes, pricing changes, or usage caveats for other readers.