MiniMax has officially brought its new-generation Flash model M3.1Flash into the spotlight. The most dramatic part is its origin — the "Space Bunny" model that previously quietly processed over 200T tokens under an anonymous identity was in fact an earlier version of M3.1Flash; in other words, before the official announcement, it had already warmed itself through and through with massive real-world traffic.
As a contender in the new-generation Flash series, M3.1Flash achieves generational-level improvements in Agentic tasks, programming, Computer Use, and cybersecurity capabilities. Even more tangible are two hard metrics: online inference speed is nearly 3 times faster, while usage costs have been cut by 70% — for developers and enterprises that live on call volume, this means the same budget buys three times the throughput, and the barrier to entry is visibly lowered. MiniMax has also left a hook: a Pro version pursuing a higher ceiling of intelligence will be launched soon. With Flash responsible for pushing "fast and cheap" to the extreme, and Pro reaching for the ceiling of "how smart can it be," the company has drawn a clearer line between its two ledgers of performance and cost.