- The model adds native multimodal capabilities and is designed to handle complex coding and software-development tasks
- The launch comes as enterprise services overtake consumer products as MiniMax’s biggest source of revenue
MiniMax on September 28 launched M3.1-Flash-Preview for public testing, adding native multimodal capabilities and a 1 million-token context window as the Chinese AI company pushes further into enterprise and developer services.
The model can handle tasks including bug fixing and full-feature development, helping developers identify problems, write code, test changes and add logic for edge cases.
MiniMax positions it as an end-to-end coding model that can take a software task from diagnosis through delivery.
Native multimodality means the model was trained from the outset to process text, images and video together, rather than adding visual capabilities to a text-only model later.
Its 1 million-token context window allows it to process very large amounts of information in a single session, potentially covering an entire code repository without breaking the task into smaller sections.
‘Space Bunny’ fuels speculation
The launch has also revived speculation around “Space Bunny,” an anonymous model that attracted attention among developers earlier this month.
Space Bunny quietly went online on September 23 and topped OpenRouter and OpenCode’s daily usage rankings during the Mid-Autumn Festival holiday.
Tokenizer tests showed similarities with MiniMax models, prompting some developers to speculate that Space Bunny could have been an early preview of M3.1-Flash.
Neither MiniMax nor the developers behind Space Bunny have publicly confirmed the connection.
Enterprise business takes the lead
The launch extends MiniMax’s shift from a consumer AI application company toward an AI infrastructure provider.
MiniMax reported $117 million in revenue in the first half of 2026, up 283% from a year earlier.
Revenue from its open platform and enterprise services surged 703%, accounting for 63% of total revenue, up from 30% a year earlier and overtaking consumer products for the first time.
The continued development of the M3 series in coding and agent capabilities has helped drive MiniMax’s developer and enterprise user base above 2 million, the company said.
M3.1-Flash-Preview is currently available through MiniMax’s Token Plan and MiniMax Code. The company has not released separate token-based pricing or public benchmark results for the preview.
Header image credit: MiniMax
