MiniMax releases M3.1 Flash, says speed is up nearly 3x and cost down 70%
In short
MiniMax released its new-generation Flash model, M3.1 Flash, on the evening of 19 October. The company said online inference speed rose by nearly 3 times and usage cost fell 70%. It also confirmed that the anonymous Space Bunny model, which had topped usage rankings on OpenRouter and OpenCode, was an early version of M3.1 Flash.
Read the full story 2 min read
MiniMax formally released its new-generation Flash model, M3.1 Flash, on the evening of 19 October, according to CLS, 36Kr, IT Home and Jiemian. The company said the model improved on agentic tasks, programming, Computer Use and cybersecurity capabilities, while online inference speed rose by nearly 3 times and usage cost fell 70%. [ 4 , 5 , 6 , 7 , 8 ]
Earlier on 19 October, MiniMax's official account posted a short video captioned "Welcome back, Space Bunny". Yicai, Jiemian and 36Kr said this was seen as an official claim to the anonymous model Space Bunny. They reported that after going online in late September, it topped daily token usage on OpenRouter and OpenCode within 3 days, with DeepSeek V4.1 Flash and GLM 5.3 Flash completing the top three. Yicai said Space Bunny was an early version of M3.1 Flash. CLS, 36Kr and IT Home later said it had processed more than 200T tokens anonymously. [ 1 , 2 , 3 , 4 , 6 , 7 ]
IT Home and Jiemian gave the prices: 1 yuan per million tokens for input, 4 yuan for output and 0.1 yuan for cache reads. Jiemian said input and output unit prices were more than 50% lower than for M3, and the cache-read price more than 70% lower. It also said M3.1 Flash uses a sparse attention architecture, with daily output speed of 200+ TPS and a median (P50) of 170+ TPS. [ 7 , 8 ]
IT Home reported that the model was live on MiniMax API, MPlan and MiniMax Code, and that weights would be opened later. It said the model reached the level of leading models in complex agent tasks, Computer Use, software development and cybersecurity, and named three benchmarks: Automation-Bench, DeepSWE and Agents' Last Exam. CLS, 36Kr, IT Home and Jiemian said MiniMax had announced a Pro version, aimed at a higher intelligence ceiling, for release soon. [ 4 , 6 , 7 , 8 ]
On 20 October, Jiemian items about exchange-traded funds mentioned the release. One said the CSI Artificial Intelligence Theme Index was down 1.40% at 09:34. Another said the ChiNext computing infrastructure index was up 0.60% at 10:25. OSChina's headline said the model ranked first among domestic models in Computer Use. [ 10 , 11 , 12 ]
Why it matters
The outlets present the release as a cut in price and a gain in speed for agent and coding use, with a Pro version promised soon. The documents give no independent test of the claims, which come from MiniMax.
Key facts
- MiniMax formally released the new-generation Flash model M3.1 Flash on the evening of 19 October. [ 4 , 5 , 6 , 7 , 8 ]
- Space Bunny, which handled more than 200T tokens anonymously, is an early version of M3.1 Flash. [ 4 , 6 , 7 ]
- MiniMax said online inference speed rose by nearly 3 times and usage cost fell 70%. [ 4 , 6 , 7 , 8 ]
- API prices are 1 yuan per million tokens for input, 4 yuan for output and 0.1 yuan for cache reads. [ 7 , 8 ]
- Jiemian reported that input and output unit prices are more than 50% lower than for M3, and cache-read price more than 70% lower. [ 8 ]
- IT Home reported that the model is live on MiniMax API, MPlan and MiniMax Code, with weights to be opened later. [ 7 ]
- MiniMax said a Pro version aimed at a higher intelligence ceiling will come soon. [ 4 , 6 , 7 , 8 ]
Confirmed by several sources
- MiniMax released M3.1 Flash on 19 October, and Space Bunny was an early version of it. [ 4 , 6 , 7 , 8 ]
- The company said M3.1 Flash improved on agentic tasks, programming, Computer Use and cybersecurity, with inference speed up nearly 3 times and cost down 70%. [ 4 , 6 , 7 , 8 ]
- Prices are 1 yuan per million input tokens, 4 yuan per million output tokens and 0.1 yuan per million cache-read tokens. [ 7 , 8 ]
Still unclear
- When the model weights will be opened. IT Home said only that weights will be opened later; no date is given.
- When the Pro version will arrive. MiniMax said only that it will come soon.
- Whether M3.1 Flash is first among domestic models in Computer Use. Only the OSChina headline says so; the other documents make no such claim.
- How the benchmark results compare with other models. IT Home relays the company's description of results on Automation-Bench, DeepSWE and ALE without giving scores.
- MiniMax's official account posted a short video captioned "Welcome back, Space Bunny" on the afternoon of 19 October, seen as claiming the anonymous model. Reported by a single source so far
What local media are saying
Timeline, local time
- Yicai reported that MiniMax's official account had claimed Space Bunny and that M3.1 Flash was about to be released. [ 1 ]
- CLS reported that MiniMax had formally released M3.1 Flash. [ 4 ]
- Yicai reported the formal release of M3.1 Flash. [ 5 ]
- Jiemian reported the architecture, output speed and price details. [ 8 ]