MiniMax Launches M3.1 Flash AI Model
Executive Summary
"MiniMax has officially unleashed its M3.1 Flash AI model, an ultra-efficient inference powerhouse that previously processed over 200 trillion tokens in secret under the alias Space Bunny. Designed to drastically slash operational overhead, this new architecture promises to redefine high-speed developer workflows."
Artificial intelligence infrastructure is shifting fast toward hyper-efficiency, and MiniMax is making a loud statement with its latest release. The company has officially rolled out the M3.1 Flash model, a high-performance system built from the ground up to tackle fast-paced inference while slashing operational overhead for developers. Before its public unveiling, this architecture quietly processed over 200 trillion tokens under the anonymous alias Space Bunny. Now, it steps into the daylight as a major contender in agentic automation and multi-step execution.
Speed and affordability define this new release. MiniMax has leaned heavily into sparse architectures to optimize output speeds, allowing the model to consistently pump out between 170 and 200 tokens per second in daily use. More importantly, the financial barrier to entry drops dramatically. Compared to its predecessor, overall usage costs for agentic operations plunge by 70 percent. Developers can now access the model through a flat pricing structure that charges just $0.15 per million input tokens, $0.60 per million output tokens, and an ultra-cheap $0.015 for cached input.
The competitive pricing does not come at the expense of raw capability. Across rigorous industry benchmarks, M3.1 Flash punches well above its weight class. It secures top-tier placement on AutomationBench-AA for end-to-end enterprise workflows and scores strongly on DeepSWE v1.1 for software engineering tasks. Furthermore, the model demonstrates impressive proficiency in computer use and cybersecurity vulnerability discovery, scoring 82 percent on CyberGym. These metrics highlight a broader industry trend where smaller, faster flash-tier models are starting to match the heavy lifting previously reserved for oversized foundational models.
Pricing and Accessibility Breakdown
- Input tokens: $0.15 per 1M
- Output tokens: $0.60 per 1M
- Cached input tokens: $0.015 per 1M
- Deployment channels: MiniMax API, M Plan, and MiniMax Code
The market impact of this release is straightforward. As enterprise adoption moves past simple chatbots and toward autonomous software agents that execute multi-step workflows across SaaS tools, inference cost becomes the ultimate bottleneck. By aggressively trimming the fat off operational budgets while retaining high-level performance in coding and tool usage, MiniMax gives startups and established engineering teams a viable engine for scaling automation without blowing past their cloud budgets.