MiniMax M3.1 Flash: Nearly 3x Faster and 3x Cheaper Than M3...

MiniMax M3.1 Flash: Nearly 3x Faster and 3x Cheaper Than M3, Weights to Be Released on October 23

On October 19, 2026, MiniMax officially released the M3.1 Flash model, optimized for agentic tasks, code, and Computer Use, featuring nearly 3x faster online inference, approximately 3x lower costs for agentic tasks, and API pricing starting at $0.15 per 1 million input tokens.

Compiled by Sergey KostenchukPublished 2026-10-19Updated 2026-10-19

2026-10-19 Coding Anthropic

🛠 MiniMax M3.1 Flash: 3x Faster and 3x Cheaper Than M3, Weights to Be Released on October 23

On October 19, MiniMax released a model for agentic tasks, code, and Computer Use. The anonymous "Space Bunny," which had been topping rankings and processing 200+ trillion tokens, turned out to be an early version. API: $0.15 per 1 million input tokens, $0.60 per output tokens, $0.015 per cache.

🌍 On Automation-Bench (62%), DeepSWE v1.1 (73.7%), and CyberGym (82%), the model performs on par with leaders at 200+ tokens/s and a price 3x lower than M3. This puts pressure on closed flagships in mass coding and automation agents, where cost per task is the deciding factor.

👤 The model is available via API, the M Plan subscription, and the desktop MiniMax Code; on OpenRouter, it is free for 72 hours. However, "minus 70%" is a vendor's averaged estimate, and a fallback to a flagship model is required for complex terminal tasks.

Source 1: https://www.minimax.io/blog/minimax-m31f Source 2: https://www.unite.ai/minimax-launches-m3-1-flash-revealing-space-bunny-stealth-model/

Sources