Available worldwide on Xiaomi’s developer site MiMo Studio, Hugging Face and the API Platform, MiMo-V2-Flash “particularly excels in reasoning, coding and agentic scenarios, while also serving as an excellent general-purpose assistant for everyday tasks”, the company said in its MiMo blog post on Tuesday.
Chinese AI prodigy Luo Fuli, who recently joined Xiaomi’s MiMo team after a stint at DeepSeek, wrote in a post on X that MiMo-V2-Flash served as “step 2 on our AGI road map”. AGI, the acronym for artificial general intelligence, refers to a hypothetical type of AI that reaches or surpasses human cognitive capabilities.
Advertisement
According to Xiaomi, MiMo-V2-Flash was engineered for maximum efficiency. “It delivers blazing-fast inference at 150 tokens per second, while maintaining an ultra-low cost of US$0.1 per million input tokens and US$0.3 per million output tokens – making it one of the most cost-effective high-performance models available,” the company said.
It claimed that MiMo-V2-Flash achieved performance comparable to Moonshot AI’s Kimi K2 Thinking and DeepSeek V3.2 Thinking on most reasoning benchmarks, while maintaining competitive general writing capabilities for high-quality open-ended responses. In long-context evaluations, Xiaomi said its latest model surpasses Kimi K2 Thinking.
Advertisement
On agentic tasks, Xiaomi touted that MiMo-V2-Flash scored 73.4 per cent on SWE-Bench Verified to outperform all open-source competitors, while approaching the performance of OpenAI’s GPT-5-High.
It was also on par with Claude 4.5 Sonnet, considered the industry’s best coding model, at a fraction of the Anthropic model’s cost.
Xiaomi launches open-source model to compete against DeepSeek, OpenAI systems
