
MiniMax Launches M3.1 Flash: Nearly 3x Faster, 70% Cheaper
MiniMax rolled out M3.1 Flash on October 19, a speed-tier model aimed at agentic tasks, coding, computer use, and cybersecurity — with API pricing at $0.15 per million input tokens and a 1M context window. Open weights are not yet dated.
Chinese AI lab MiniMax launched M3.1 Flash on Monday, October 19, an updated version of its flagship speed-tier model. Reuters reported the launch from Beijing. The company says the model delivers what it calls a “generational leap” in agentic tasks, coding, Computer Use, and cybersecurity, with online inference running nearly three times faster than its predecessor while cutting usage costs by 70%. Those figures come from MiniMax’s own release materials and have not been independently verified.
M3.1 Flash is available through three channels: the MiniMax API, the company’s M Plan subscription, and MiniMax Code, its developer-facing coding agent tool. API pricing is set at $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cached-input tokens — firmly in the budget tier of commercial LLM pricing. The model ships with a 1 million token context window, enough to hold an entire codebase, a long chat history, or a multi-document research session in a single pass.
What makes the rollout notable is the path it took to get here. A preview version, M3.1-Flash-Preview, had already appeared inside MiniMax Code on September 27, roughly three weeks before the formal announcement — with no public model card, no benchmark report, and not even pricing information. Earlier still, in late September, an anonymous free listing called “Space Bunny Alpha” showed up on OpenRouter. Community researchers noticed that its outputs, latency profile, and coding behavior lined up closely with what would become M3.1 Flash. MiniMax later confirmed the connection itself, posting on its official account that Space Bunny had “topped global leaderboards with 200T+ tokens” and was, in fact, an earlier version of M3.1 Flash.
This kind of stealth testing — releasing a model anonymously to gather real-world usage data and leaderboard standing before attaching a brand name and a price tag — has become increasingly common among frontier labs. It lets a lab tune against live traffic and build word-of-mouth among developers without spending a dollar on marketing. MiniMax is just unusually open about having done it.
There is no confirmed date for open weights. MiniMax has open-weighted earlier releases, but the company did not commit to a timeline on launch day. The price sheet, by contrast, arrived immediately — a signal that this release is aimed squarely at paying API customers rather than the open-source community.
The timing puts M3.1 Flash in the middle of an already crowded year for Chinese AI labs. MiniMax, DeepSeek, Zhipu, Alibaba’s Qwen team, and Moonshot AI’s Kimi have all been racing to ship faster, cheaper models aimed at the same audience: developers building agents and coding tools, who care more about latency and token price than raw benchmark bragging rights. M3.1 Flash is MiniMax’s latest entry in that race — and at $0.15 per million input tokens, it is priced to be hard to ignore.
A few things are still missing from the picture. MiniMax has not published a benchmark report or a model card for M3.1 Flash, and the preview that shipped inside MiniMax Code in late September arrived without either — unusual for a model the company now markets as a flagship-tier upgrade. The company also did not say when, or whether, open weights will follow, despite having open-weighted earlier releases. For now, the verifiable substance of the launch is the product surface: three access channels, a published price sheet, a 1M context window, and the company’s own performance claims. Independent evaluation will have to wait until researchers can put the released model through its paces.
If the “generational leap” claim holds up under independent testing, M3.1 Flash would sharpen a trend that has defined 2026 for Chinese labs: competing on speed and price for the agent-building crowd rather than on headline benchmark scores. If it doesn’t, MiniMax still walks away with something valuable — weeks of real-world traffic data gathered under the Space Bunny alias, and a pricing sheet that undercuts most of the commercial field. Either way, the launch is a data point worth watching in the fastest-moving segment of the model market.