STORY · MODELLER_
MiniMax-M2.5 open source with Agent focus – performs on SWE-Bench and cheap inference
MiniMax has released M2.5 as open source with Forge, a reinforcement learning framework trained on over 200,000 RL environments for coding, tool calling, and agents. The model achieves 80.2% on SWE-Bench Verified and is marketed as cost-effective with inference costs around 1 dollar per hour at 100 tokens per second.
WHY IT MATTERS
This is a significant step for open-source models in agent capabilities and code generation, and the price point makes it practical for production. Competition with Together and GLM-5 indicates that open-source models are now catching up to proprietary alternatives for complex workflows.
SOURCES
MACHINE-GENERATED SUMMARY This summary is written by machine from the sources below. We sort and explain — but we are a way into the field, not the final word. Check the source when something matters to you.