MiniMax M3.1 Flash Rolls Out, Cuts Costs by 70%

Chinese AI company MiniMax has officially rolled out its next-generation model, M3.1 Flash, which is designed for agentic tasks, coding, computer use, and cybersecurity workloads. The announcement was made on October 19, 2026.

M3.1 Flash is the latest addition to MiniMax’s ‘Flash’ line of models, which are known for their speed and efficiency. According to the company, M3.1 Flash is built for agentic tasks such as coding, computer use, and cybersecurity workloads.

The model has already been tested in real-world scenarios under a pseudonym called ‘Space Bunny’, which processed over 200 trillion tokens before being officially named and launched by MiniMax. This stealth-testing approach has become increasingly common in the AI industry, where labs quietly route unnamed or pseudonymous models through public leaderboards and community coding tools to gather feedback at scale.

M3.1 Flash’s performance is impressive, with benchmark scores showing it excels in areas such as DeepSWE (74%), Automation-Bench (62%), ALE (30%), and BabyVision (91%). These numbers indicate that the model is strong in agentic task completion, coding, visual reasoning, and other key areas.

The pricing for M3.1 Flash has been significantly reduced compared to its predecessor, with a 70% cut in usage costs. The company’s official API pricing stands at $0.15 per 1 million input tokens and $0.60 per 1 million output tokens, with cached reads priced separately at $0.015 per 1 million tokens.

M3.1 Flash is now available through three channels: the MiniMax API for developers building directly against the model, the M Plan subscription tier for higher-volume users, and MiniMax Code, the company’s dedicated coding-agent product. This marks a significant milestone in the development of AI models, as companies like MiniMax continue to push the boundaries of what is possible with agentic workloads.

The release of M3.1 Flash has sparked interest among developers and researchers alike, who are eager to test its capabilities and explore new applications for this technology. As the industry continues to evolve, it will be fascinating to see how models like M3.1 Flash shape the future of AI development.

MiniMax’s decision to release model weights on October 23, 2026, four days after the initial announcement, is a common practice in the industry. This allows the company time to finalize documentation and safety testing before making the weights available to the open-source community.

The benchmark scores for M3.1 Flash are impressive, but it’s essential to note that direct comparisons across different models can be challenging due to varying benchmark suites and scoring systems. However, one thing is clear: every major lab in this tier is competing on speed, price, and performance.

MiniMax has built a reputation as an aggressive player in the Chinese AI market, with successive Flash and standard-tier models shipped at a rapid pace throughout 2026. The company’s focus on pricing and iteration speed has made it a recurring name in industry conversations alongside other major players like DeepSeek and Moonshot AI.

The stealth-model playbook adopted by MiniMax is not unique to this company; many labs have followed suit, treating the earliest phase of a model’s life as a private field test rather than a tightly controlled internal benchmark exercise. This shift has been driven partly by the rise of public model-routing platforms and community coding tools.

MiniMax’s announcement has sparked interest among enterprise buyers outside China who are evaluating whether to add Chinese-developed models like M3.1 Flash to their infrastructure. The company’s credibility with these customers is built on consistent messaging, delivered through state-recognized wire services such as Reuters.

The release of model weights for M3.1 Flash will allow independent researchers and infrastructure providers to benchmark the model on their own hardware, verify MiniMax’s speed claims under different serving stacks, and potentially offer hosted versions at their own pricing. This verification step is a crucial aspect of the AI open-source community’s push for transparency and accountability in model development.

M3.1 Flash joins an increasingly crowded market where every major lab now ships a fast, cheap ‘second tier’ model alongside its flagship. The competitive landscape has become more intense as companies focus on pricing, speed, and performance to differentiate themselves from rivals.

The release of M3.1 Flash marks the culmination of MiniMax’s efforts in 2026 to build credibility with enterprise buyers outside China who are evaluating whether to add Chinese-developed models like this one to their infrastructure.