Artificial Analysis reports that MiniMax-M3 offers a 1M-token context window and competitive API pricing, while MiniMax says the model combines coding, agentic capabilities, native multimodality, and an open-weight release plan.
Artificial Analysis reported that MiniMax-M3 is a new model with a 1M-token context window, published API pricing, and benchmark results that position it prominently among open-weight models once its weights are released.
MiniMax describes M3 as an open-weight model built for coding, agentic tasks, long-context use, and native multimodality. On its product page, MiniMax lists the API model name as MiniMax-M3 and says the model supports a 1M-token MSA context window.
In a company blog post, MiniMax says M3 was released on June 1, 2026, and calls it the first open-weight model to combine “frontier coding,” 1M context, and native multimodality in one model. The same MiniMax post says the model’s technical report and weights will be released over the next 10 days, which means independent users should still watch for the actual availability of those materials before treating the open-weight release as complete.
Artificial Analysis reports MiniMax-M3 pricing at $0.30 per 1 million input tokens and $1.20 per 1 million output tokens for prompts up to 512K context. For usage from 512K to 1M context, Artificial Analysis lists pricing at $0.60 per 1 million input tokens and $2.40 per 1 million output tokens.
Those figures matter because long-context models can become expensive when users process large documents, codebases, or multi-turn agent histories. MiniMax’s stated 1M-token context window, if it performs reliably in production, gives developers room to test workflows that require unusually large context inputs.
Artificial Analysis says MiniMax-M3 reaches an Intelligence Index score of 55 and reports a GDPval-AA result of about 1670. The publication frames MiniMax-M3 as a leading open-weights model once the weights are released, tying that assessment to its benchmark comparisons and pricing data.
As with all benchmark claims, the details of test methodology and real-world fit matter. Artificial Analysis provides comparative benchmark reporting, while MiniMax provides the product positioning and release claims. Developers evaluating M3 should compare those external results with their own workloads, especially for long-context retrieval, coding tasks, tool use, and multimodal inputs.
The announcement is notable for the combination of claims rather than any single feature. MiniMax is presenting M3 as an open-weight model with native multimodality, coding and agentic capabilities, and a 1M-token context window. Artificial Analysis adds pricing and benchmark comparisons that suggest the model could be cost-competitive for large-context use.
The key caveat is timing. MiniMax says the technical report and weights will be released over the next 10 days from its June 1, 2026 announcement. Until those weights and technical details are available, outside developers and researchers cannot fully verify the open-weight characteristics or reproduce the company’s claims.
For now, MiniMax-M3 is best understood as a newly announced long-context model with published API pricing, company-stated multimodal and coding capabilities, and early third-party benchmark coverage from Artificial Analysis.
What MiniMax is claiming MiniMax describes M3 as an open weight model built for coding, agentic tasks, long context use, and native multimodality.
On its product page, MiniMax lists the API model name as MiniMax M3 and says the model supports a 1M token MSA context window.
In a company blog post, MiniMax says M3 was released on June 1, 2026, and calls it the first open weight model to combine “frontier coding,” 1M context, and native multimodality in one model.
Continue reading
