MiniMax M3.1 Flash Drops Agent Costs
MiniMax’s new M3.1 Flash targets agentic workflows with a 1M-token context window, 200+ tokens-per-second output, and strong results across automation, coding, computer-use, and cybersecurity benchmarks. At $0.15 per million input tokens and $0.60 output, it is available through the MiniMax API, M Plan, and MiniMax Code, with weights to follow. [MiniMax announcement](https://www.minimax.io/blog/minimax-m31f)
M3.1 Flash’s real pitch is making long-running agents cheap enough to operate continuously, not merely winning another benchmark. The main caveat is that its headline comparisons remain vendor-reported, so developers should validate tool reliability and task completion on their own workloads.
- –The 1M-token context window is well suited to large repositories, extended agent sessions, and document-heavy workflows.
- –$0.15 input and $0.60 output pricing materially lowers the cost of high-volume agent execution.
- –Claimed 200+ TPS peak output could make interactive coding and computer-use agents feel substantially more responsive.
- –AutomationBench, BankerToolBench, DeepSWE, and OSWorld results suggest broad utility beyond coding alone.
- –The upcoming weight release could turn M3.1 Flash from a hosted bargain into a compelling local or self-hosted option.
DISCOVERED
1h ago
2026-10-19
PUBLISHED
1h ago
2026-10-19
RELEVANCE
AUTHOR
0x0SojalSec
