Key Info

MiniMax announced M3.1 Flash, describing it as the earlier version behind Space Bunny, a model that topped global leaderboards with over 200T tokens processed. The company says it delivers frontier-level intelligence at one-third the cost of M3 and three times the speed.

Highlights

  • Benchmark results cited: Automation-Bench 62% and ALE 30% for Cowork tasks, DeepSWE 74% for coding, BabyVision 91% for visual reasoning
  • Flat-rate API pricing: $0.15 per 1M input tokens and $0.60 per 1M output tokens, with cache reads at $0.015 per 1M tokens
  • Model weights are scheduled to arrive on Oct 23
  • Positioned as redefining the performance economics of agentic AI