MiniMax Launches M3.1 Flash: 3x Faster Inference, 70% Cheaper, Top Agentic Benchmarks
MiniMax 稀宇科技 · wechat · 2026-10-19
MiniMax officially released MiniMax M3.1 Flash, revealing that SpaceBunny—which anonymously served 200T+ tokens—was an early version of it.
- Capabilities: top-tier results across agentic tasks (Automation-Bench-AA), Computer Use (OSWorld V2: 61.7, ALE: 29.6, both #1 among Chinese models), coding (DeepSWE v1.1: 73.7%), and cybersecurity (CyberGym: 82%).
- Performance & price: aggressive sparse attention delivers 200+ TPS output (P50 170+ TPS); agentic task cost cut 70% vs predecessor—¥1/M input tokens, ¥4/M output, ¥0.1/M cache reads.
- Availability: live on MiniMax API, MPlan, and MiniMax Code desktop; open weights coming later, with a higher-capability Pro model teased.
Related event: MiniMax Releases M3.1 Flash: 3x Faster, One-Third the Cost of M3(26 posts)→
More from Models
- MiniMax confirms Space Bunny was early M3.1 Flash: $0.15/M input, open weights Oct 23 — lmoroney · 2026-10-20
- Meta Muse hits 5M downloads in 22 days, beating ChatGPT's pace — aakashgupta · 2026-10-20
- BioEmu2 tease: successor to the protein dynamics model may drop soon — sokrypton · 2026-10-20
- Gemini 4 Argon Beats Claude Opus 5.5 at 2026 ICPC Xi'an Regional — boker_tov · 2026-10-20
- OmicsBench v1: 784 omics tasks benchmark 31 model-harness combos, frontier models tie — kenbwork · 2026-10-20
- Gemini turns a minimal sketch into a full working humanoid robot control sim — bousmalis · 2026-10-20