MiniMax M3.1 Flash vs. MiniMax M3: One is cheaper, but not faster or better

In a recent comparison, MiniMax M3.1 Flash emerged as a cost-effective alternative to the MiniMax M3, though its performance gains are nuanced. MiniMax introduced M3.1 Flash on October 19, 2026, claiming it reduces the overall expense of agentic tasks by 70%.

Flash is priced at $0.15 per million input tokens and $0.60 per million output tokens. These rates are half the discounted pricing of M3, which stands at $0.30 and $1.20 for requests under 512K input tokens; separate rates apply for cached inputs and longer M3 requests.

Independent scoring by Artificial Analysis noted M3.1 Flash achieved 72.5% on AutomationBench-AA, a measure of objective completion without guardrail breaches in simulated business applications. This differs from Zapier’s AutomationBench leaderboard percentage for fully completed tasks.

When testing three smaller agent tasks, the results showed distinct operational styles. In the agentic bug fix scenario, Flash successfully resolved all three bugs but exhausted its turn limit before submitting a final answer. M3 identified the wrong test but missed one instance of the second bug.

For the incident investigation, both models achieved a perfect 10 out of 10 score, successfully identifying the config change, ruling out decoy deploys, and noting the slow clock. Flash required 17 turns, bundling more calls per turn, while M3 needed 24 turns. Flash demonstrated an estimated cost about 68% lower on this test, aligning closely with MiniMax’s 70% claim for agentic tasks.

In the long context test, neither model yielded a final answer. Flash utilized the 64,000 output tokens allowed in the initial run and stopped, and M3 also ceased after approximately 29,000 output tokens. Despite MiniMax listing a roughly 1-million-token context window for M3, this capacity did not translate to a completed response in this specific test.

Overall, while M3 was faster across the board, Flash held a lower estimated total cost, even when consuming more input and output tokens. The author concludes that apart from the lower estimated cost, the tests did not demonstrate clear performance improvements over M3.