Minimax/Minimax-M2.5

MiniMax officially launched MiniMax-M2.5 on February 12, 2026, positioning it as a frontier model built for real-world productivity in coding, agentic tool use, search, and office work. The model was extensively trained with reinforcement learning in hundreds of thousands of complex real-world environments. M2.5 is the first frontier model where users do not need to worry about cost, delivering on the promise of intelligence too cheap to meter. M2.5 achieves 80.2% on SWE-Bench Verified, 51.3% on Multi-SWE-Bench, and 76.3% on BrowseComp, completing the SWE-Bench Verified evaluation 37% faster than M2.1 and matching the speed of Claude Opus 4.6. It costs just $1 to run the model continuously for an hour at 100 tokens per second, dropping to $0.30 at 50 TPS. The model also exhibits a spec-writing tendency, planning project architecture before writing code.

MiniMax-M2.5 is out: 204,800 context, 50/100 TPS, strong SWE-Bench Verified + tool/search results, RL system details (Forge), and API integration notes.

Third-party aggregator Magica lists M2.5 with a 204.8K token input context window and 128K token output limit, released February 12, 2026. The model supports function calling, structured output, and reasoning mode (always enabled). M2.5 is also available on HuggingFace as an open-source release, according to the aggregator's listing. Magica reports M2.5 pricing at $0.27 per million input tokens and $1.08 per million output tokens, with a cache read cost of $0.03 per million tokens. The model achieved a Best Design Arena Score of 1,230 Elo with a 57.5% win rate in the Website subcategory. Magica positions M2.5 for comparison against models including GLM 5.3 Flash, DeepSeek V4.1 Flash, and MiMo-V2.6-Flash in programming tasks.

Per OpenRouter's model listing, MiniMax M2.5 is a SOTA large language model released on February 12, 2026, with text-in/text-out modalities and a 205K-token context window. It is positioned for real-world productivity, extending the coding strengths of M2.1 into general office work such as generating and operating Word OpenRouter reports benchmark scores of 80.2% on SWE-Bench Verified, 51.3% on Multi-SWE-Bench, and 76.3% on BrowseComp, and lists listed pricing of $0.22 per 1M input tokens and $0.90 per 1M output tokens, with weighted-average effective prices of about $0.0866 input and $0.9901 output per 1M tokens. The page enumerates