MiniMax M3 goes open-weight with 1M-token context, beats GPT-5.5
First open-weight model with 1M-token context and native computer use capabilities.
MiniMax has released M3, the first open-weight model to combine a 1-million-token context window with native multi-modal computer use capabilities. Scoring 59% on the SWE-Bench Pro benchmark, M3 outperforms both GPT-5.5 and Gemini 3.1 Pro, establishing a new frontier for open-weight models in software engineering and agentic tasks. The model can process entire codebases, interact with GUIs, and perform multi-step reasoning without external tools, making it a strong candidate for enterprise automation.
The open-weight release allows developers to self-host M3, avoiding API costs and data privacy concerns. With its long context and computer-use abilities, M3 enables applications like automated bug fixing across large repositories, end-to-end UI testing, and real-time document summarization for legal or financial workflows. This positions MiniMax as a key player in the open-source AI race, challenging major closed-source models while offering transparency and customization.
- M3 achieves 59% on SWE-Bench Pro, beating GPT-5.5 and Gemini 3.1 Pro
- First open-weight model with native multi-modal computer use and 1M-token context window
- Self-hostable, reducing API costs and improving data privacy for enterprises
Why It Matters
Open-weight models now rival top proprietary systems, enabling affordable, private, and customizable AI for complex enterprise workflows.