MiniMax officially launches M3.1 Flash, reveals Space Bunny as its early version, publishes $0.15/$0.60 pay-go pricing and agentic benchmarks; open weights promised later

Why it matters

【Key points】

On October 19 (Beijing time evening), MiniMax formally launched M3.1 Flash and confirmed that Space Bunny 🐰 — the anonymous model that processed 200T+ tokens on global leaderboards — was an earlier version of M3.1 Flash. Official post: "MiniMax M3.1 Flash: Redefining Agent Performance and Cost"; @MiniMax_AI posted ~19:56 CST.

Pay-go pricing: $0.15 input / $0.60 output / $0.015 cached read per 1M tokens. MiniMax claims ~70% lower agentic usage cost vs the prior generation, nearly 3× faster online inference, and 200+ TPS (P50 ~170+). Available via MiniMax API, M Plan and MiniMax Code; OpenRouter lists minimax/minimax-m3.1-flash:free. Blog: "Model weights will be released later" — HF MiniMaxAI still shows Music3/H3/M3 only, no M3.1 Flash weights yet.

Self-reported scores: top-tier AutomationBench-AA; BankerToolBench 76.2; DeepSWE v1.1 73.7%; OSWorld V2 61.7; ALE 29.6 (first among Chinese models per MiniMax); CyberGym 82%. A capability-focused Pro is teased for later. Same-day e/952 covered docs-only API GA + M2.5/M2.7 retirement, not this launch post, pricing, benchmarks or Space Bunny reveal.

【Verifiable】

Blog: https://www.minimax.io/blog/minimax-m31f

Release-notes card (#### Oct. 19, 2026): https://platform.minimax.io/docs/release-notes/models.md

Pay-go pricing: https://platform.minimax.io/docs/guides/pricing-paygo.md

@MiniMax_AI: https://x.com/MiniMax_AI/status/2112150867868541221

【Our read】

1. The real public launch after e/952's docs GA: blog, official post, price and benches, plus Space Bunny unmasked. Weights still pending.

2. $0.15/$0.60 is close to earlier $0.165/$0.66 rumors; OpenCode "free for N days" is a channel claim — trust official + channel pages.

3. Benchmarks are vendor-reported; independent runs needed. Pro is teased with no date.