MiniMax Adds M3.1-Flash-Preview Coding Model to MiniMax Code
The AMW Read
An in-product coding model expands MiniMax's offering, but missing benchmarks, API access and pricing limit what this launch establishes about its competitive impact.
MiniMax Adds M3.1-Flash-Preview Coding Model to MiniMax Code
MiniMax introduced M3.1-Flash-Preview inside its MiniMax Code agent tool on September 27, 2026. The coding model supports a context window of up to 1 million tokens and offers five reasoning-effort settings: low, medium, high, xhigh and max. At launch, MiniMax had published no model card, benchmark report, public API listing or price for M3.1-Flash-Preview. That leaves its performance and cost outside MiniMax Code unverified.
The release puts a new foundation model directly into a developer workflow, where users can assess it on real coding tasks before MiniMax offers a standalone API. That distribution matters in a market where model labs compete for developer use as well as benchmark standing. The article links the release to an anonymous OpenRouter model called Space Bunny Alpha through community fingerprinting tests, but MiniMax has not confirmed they are the same system. Its reported pricing for the earlier M3 model also provides no basis for assuming how M3.1-Flash-Preview will be priced.
Builders evaluating MiniMax Code can test the new model on their own bug fixes and feature work, measuring completion quality and latency at each effort setting. Teams considering it for production should establish whether the agent tool fits their workflow; the absence of a public API and published pricing limits what they can yet conclude about direct integration or operating cost. For investors, adoption inside MiniMax Code is the near-term signal to watch, while independent performance and pricing evidence remain outstanding.
