MiniMax unveils M2.5 model

MiniMax has launched M2.5, a foundation model it said delivers state-of-the-art coding and agentic performance, including 80.2% on the SWE-Bench Verified benchmark. Served at up to 100 tokens per second (TPS), the model can reportedly run continuously for about USD 1 per hour at 100 output tokens per second, with a lower-speed version priced at roughly half the cost.