MiniMax launches M3.1 Flash, the model behind the anonymous 'Space Bunny'

Chinese AI lab MiniMax released M3.1 Flash on 19 October 2026, a multimodal model with a 1M-token context window priced at $0.15 per million input tokens [1][2]. MiniMax claims nearly three times faster inference and 70% lower usage costs [1][4]. It says the model weights will be released later [1].

What is M3.1 Flash?

MiniMax launched M3.1 Flash on 19 October 2026 at $0.15 per million input tokens and $0.60 per million output tokens. Cached input costs $0.015 per million, and the model is available through the MiniMax API, M Plan and MiniMax Code [1].

MiniMax's API docs list a 1M context window and tunable thinking depth [2]. OpenRouter says the model accepts text, images and video as input and returns text, with up to 524,288 completion tokens [3].

MiniMax says Space Bunny, an anonymous model that processed over 200 trillion tokens, was an early M3.1 Flash. Temperature2 reports that this build ran on OpenRouter under the codename [1][6].

How good is it, according to MiniMax?

MiniMax reports 73.7% on DeepSWE v1.1, 61.7 points on OSWorld V2 and 82% on CyberGym [1]. It claims first place among Chinese models on OSWorld V2 and Agents' Last Exam, where it scored 29.6 points [1].

The company says median output speed is consistently above 170 tokens per second, reaching 200+ in everyday use [1]. These are MiniMax's own numbers and have not been independently reproduced.

What do independent tests show?

Artificial Analysis scores M3.1 Flash at 42 on its Intelligence Index, up from 29 for M3, The Decoder reports [5]. That ties GLM-5.3-Flash, edges Gemini 3.8 Flash at 41 and trails DeepSeek V4.1 Flash 1017 at 45 [5].

Artificial Analysis measured a 93 percent hallucination rate for M3.1 Flash, according to The Decoder. The model tends to answer every question instead of declining when unsure, while cost per task is about $0.22 [5].

When do the open weights arrive?

MiniMax's blog says only that weights will be released later, and OpenRouter lists open weights as planned [1][3]. The Decoder and Temperature2 report a release date of 23 October, and Temperature2 notes that no licence has been published yet [5][6].

Our read

Opinion: At fifteen cents per million input tokens, M3.1 Flash keeps pushing China's price war toward near-free agent work. But a 93 percent hallucination rate is a real warning for unattended agents, and until the weights and licence ship, this is a cheap closed API, not an open model.

What we could not verify

• The 23 October weights date and the licence terms; MiniMax's own blog only says weights come later.
• Video input: OpenRouter lists it, MiniMax's docs say only multimodal, and The Decoder mentions text and images.
• Which predecessor the 70% cost cut is measured against; MiniMax does not name it in the blog post.
• Where exactly Space Bunny ran anonymously; MiniMax does not say, and the OpenRouter detail comes from Temperature2.
• MiniMax's benchmark scores have not been independently reproduced, apart from the Artificial Analysis results.