MiniMax has opened the weights for M3.1 Flash, just four days after the model launched on October 19, 2026. The AI company is pairing the release with production inference through Together AI, giving developers a hosted route to run the model without managing their own GPUs.
The move follows a familiar playbook: launch the model behind the company’s own API first, then widen access with downloadable weights for the open-source community. With weights in hand, developers can self-host, fine-tune and experiment — while Together AI’s hosted inference offers a production-ready path for teams that would rather skip the infrastructure work entirely.
Open-weights releases matter because they feed an entire ecosystem of tools. Runtimes like Ollama make it easy to run open models locally, and hubs like Hugging Face give them distribution — two ingredients that often decide whether a new model actually gets adopted by developers.
M3.1 Flash arrives in a crowded field of open models. Whether it wins developer mindshare will likely come down to how easily it runs on real hardware — and the Together AI partnership suggests MiniMax is thinking about exactly that.
Source: Runtime Wire
