MiniMax launches M3.1 Flash, unmasks Space Bunny model

Chinese AI startup MiniMax released M3.1 Flash on Monday, Oct. 19, and confirmed that Space Bunny, an anonymous model that had drawn heavy developer use for weeks, was an early version of it. In a statement, the company said the model brings "generational" improvements in agentic tasks, programming and cybersecurity. It also said online inference speed has nearly tripled and usage costs have dropped 70%, Reuters reported.reuters.com

The launch ends a guessing game among developers. MiniMax teased the reveal on X with the post "Welcome back. Space Bunny. MiniMax-M3.1-Flash is coming". According to the company, Space Bunny processed more than 200 trillion tokens while operating in stealth.x.com+1

Pricing and availability

API pricing is a flat $0.15 per million input tokens and $0.60 per million output tokens. Cached reads cost $0.015 per million tokens. MiniMax says the model runs at one-third the cost of its M3 model and three times the speed. M3.1 Flash is available through the MiniMax API, the M Plan subscription and the MiniMax Code desktop app. OpenRouter is offering it free for 72 hours and will then charge the same rates as MiniMax's own API. Coding tool OpenCode said Space Bunny ranked first on its platform for 12 straight days and served 106 trillion tokens. OpenCode is also offering the model free for three days.pc.watch.impress.co.jp+2

The model accepts multimodal input and has a 1-million-token context window. MiniMax says the open weights will be released on Oct. 23. The parameter count and license have not been disclosed.pc.watch.impress.co.jp

Benchmarks and stealth run

In MiniMax's own comparison table, M3.1 Flash scored 62.2% on Automation-Bench. That placed it ahead of DeepSeek-V4.1-Flash-1017 at 60.5% and Claude Opus 5 at 50.3%. It trailed DeepSeek on the DeepSWE coding test, 73.7% to 77.0%, and trailed Claude Opus 5 on the BabyVision visual-reasoning test, 91.0% to 94.1%.pc.watch.impress.co.jp

Space Bunny Alpha appeared on OpenRouter on Sept. 23 with no named developer. By Sept. 24 it led OpenCode's daily token-usage ranking. OpenRouter said MiniMax improved the model during the stealth run: time to first token fell from 1.95 to 1.22 seconds, throughput rose from 70 to 100 tokens per second, and the error rate fell from 1.4% to 0.7%.tokenpost.com+1

Pro model ahead

MiniMax said a higher-end Pro version is coming later and described it as "pushing the boundaries of intelligence." The company called Flash part of its ongoing effort to balance efficiency against capability.unite.ai