Overview
- MiniMax publicly released M3.1 Flash on Monday and billed it as nearly three times faster with a one‑million‑token context window and median output speeds above 170 tokens per second.
- The company published flat pricing of $0.15 per 1M input tokens, $0.60 per 1M output tokens and $0.015 per 1M cached reads to support its claim of much lower agent costs.
- Independent tests from Artificial Analysis and reporters found lower estimated per‑task costs in some workloads (about $0.22 versus $0.51 on the prior model) but flagged accuracy problems: the model answered 56% of factual questions correctly and usually produced wrong answers rather than abstaining.
- Key deployment details are still missing at launch because MiniMax has not published model weights, final license terms or a full description of its sparsity method, which limits enterprise auditing and self‑hosting plans.
- The M3.1 Flash rollout joins a crowded low‑cost ‘Flash’ market and could cut bills for bounded, easily verified tasks, yet real savings depend on each agent’s token mix, cache reuse, tool calls, retries and the human work needed to verify outputs.