MiniMax M3 appeared on X on Monday as the company described it as the “first open-weights model to combine three frontier capabilities,” with claims around coding and agentic performance, 1M context via “MiniMax Sparse Attention,” and “natively multimodal” support from the start.
In its announcement, MiniMax highlighted benchmark results including 59.0% on SWE-Bench Pro, 66.0% on Terminal Bench 2.1, 34.8% on SWE-fficiency, 28.8% on KernelBench Hard and 74.2% on MCP Atlas. The firm also posted a comparison graphic that placed M3 against Opus 4.7, GPT 5.5 and Gemini 3.1 Pro across a wider set of tests, though those figures should be treated cautiously as company-provided numbers.
A pricing card shared with the launch lists MiniMax M3 at $0.60 per million input tokens and $2.40 per million output tokens for “≤512K,” with prompt caching read priced at $0.12 per million tokens. For the 512K–1M tier, the posted table shows $1.20 input, $4.80 output and $0.24 for prompt caching read. MiniMax also advertised “50% off standard usage” for the first seven days on ≤512K context, with priority access via email and self-serve access “in the next few days.”
The company posted links for the API, token plan and a new MiniMax Code product, and it said weights and a tech report should arrive in about 10 days. A separate visual from the launch pages also shows M3 selected in an OpenCode Zen model menu, with a tooltip listing “text, image, video,” “Allows reasoning,” and a “Context limit 200,000.”
Source: MiniMax on X