On September 27, 2026, MiniMax confirmed through its official account that M3.1-Flash-Preview, a new text model, is now live inside MiniMax Code and selectable in the company's coding assistant. The rollout could not look more different from the high-profile M3 launch in June, which arrived with architecture details, benchmark scores, and a pricing sheet: this time MiniMax published no model card, no benchmark data, and no price — developers had already spotted the model in the MiniMax Code model picker a day earlier, and the company only confirmed it afterward.
What happened: product-only for now, differentiated by reasoning tiers
M3.1-Flash-Preview can currently be used only through MiniMax Code; its API endpoint remains gated. According to the company, the model targets everyday development scenarios with an emphasis on speed and stability, covering tasks from quick bug fixes to full feature development. The picker now offers multiple reasoning levels from low up to a new "max" tier, sitting alongside M3, M2.7, and M2.7-highspeed.
The launch comes with two limited-time perks: from September 28 to October 7 (UTC+8), new and existing users can check in daily to receive double free credits, usable on M3.1-Flash-Preview as well as H3 and H3 Max; Token Plan quotas are reset for all users at launch, with more resets promised to follow.
Background: the buzz M3 created needs a volume product line to follow
M3 was MiniMax's flagship move this June: a self-developed sparse-attention architecture, native 1M context, beating GPT-5.5 and Gemini 3.1 Pro on SWE-Bench Pro, with API pricing aimed squarely at the low-cost tier. On its mid-year earnings call in August, management said M3.1 would ship in the second half of 2026. M3.1-Flash-Preview delivers on that timeline — but in a surprising form: not a flagship iteration, rather a Flash tier built to absorb everyday coding workloads.
A quiet rollout like this is not unusual: let real users generate volume inside the company's own product first, gather feedback, harden stability, and only then decide whether to open it up via the public API. For model vendors, the Flash tier was never about leaderboard positions — it is about usability per unit of cost. High-frequency, low-complexity tasks like bug fixes, test writing, and comment generation are precisely where coding assistants burn the most tokens.
Impact: good news for MiniMax Code users; API developers will have to wait
For existing MiniMax Code users, this is the most direct win: a faster, cheaper everyday option, plus credit perks. For developers on MiniMax's pay-as-you-go API, nothing changes for now — M3.1-Flash-Preview has not entered the public API, and the only way to try it is inside MiniMax Code.
Zooming out, the coding-model race is shifting from flagship head-to-heads to tiered lineups: M3 sets the capability ceiling while the Flash tier soaks up daily traffic. The absence of benchmarks and pricing for M3.1-Flash-Preview suggests MiniMax is in no hurry to enter public comparisons — it is validating the model as a product feature first. That release cadence is itself a signal worth watching.
