MiniMax M3 is now available on Vercel AI Gateway. For product engineers already routing models through the gateway, this is a straightforward addition with some meaningful capabilities behind it.
M3 is MiniMax's first model to combine a 1M-token context window with native multimodality. That context size matters for use cases where you need to reason over large codebases, long conversation histories, or extensive documents without chunking. The model is built around MiniMax Sparse Attention (MSA), the architectural choice that makes that window size practical.
On the capability side, MiniMax reports that M3 improves on software engineering, terminal-based tool use, and agentic web browsing. It is also tuned for multi-turn collaboration, which points toward agent workflows where the model needs to stay coherent across many exchanges rather than treating each turn as a fresh context.
Integration through the Vercel AI SDK is direct. Set the model field to minimax/minimax-m3 and you are routing to M3. To use its multimodal capabilities, pass an image alongside your text input. No additional configuration is described beyond that.
The pairing with Vercel AI Gateway means you get M3 through the same interface you may already use for other models. Switching between models or testing M3 against your current setup does not require rebuilding how you handle API calls.
If you are building anything that involves long context, multi-step agent loops, or tool use over a terminal, M3 is worth testing today. Swap in the model string, run your existing evals, and see how it holds up on the tasks that matter for your product.