MiMo-V2-Flash: Open-Source LLM Redefines Reasoning and Coding

MiMo-V2-Flash, the new open-source LLM, excels in reasoning, coding, and agentic tasks. Boasting 309B parameters, it rivals top models on benchmarks like AIME 2025. Featuring a hybrid attention mechanism and Multi-Token Prediction, it achieves 150 tokens/second at $0.1/$0.3 per million input/output tokens. It also scores 73.4% on SWE-Bench Verified, rivaling GPT-5-High.