Chinese AI Company MiniMax Releases Open Model M3
MiniMax M3 is the first open model to combine frontier coding performance, a 1 million token context window, and native multimodality, directly challenging the capabilities of leading closed-source models.
Reporting from 1 sources: GIGAZINE.
Chinese AI company MiniMax has announced the open model MiniMax M3, which it says matches or exceeds the benchmark scores of GPT-5.5 and Claude Opus 4.7 in coding and agent tasks. The model supports a 1 million token context window and multimodal input, and will be released as an open model within 10 days. API and subscription pricing have also been announced.
Shanghai-based AI company MiniMax announced the open model MiniMax M3, claiming benchmark scores that compete with GPT-5.5 and Claude Opus 4.7. The model uses a sparse attention architecture that reduces computation for 1 million tokens to one-twentieth of the previous generation, with speed improvements of over 9 times in prefilling and over 15 times in decoding. On the SWE Bench Pro coding benchmark, M3 surpassed GPT-5.5 and Gemini 3.1 Pro; on SVG Bench, it outperformed Claude Opus 4.7. MiniMax stated the model will be distributed as an open model within 10 days of the announcement.
API pricing is set at $0.60 per million input tokens, $2.40 per million output tokens, and $0.12 per million cache reads for up to 512,000 tokens, with higher rates for longer contexts. Flat-rate subscriptions range from $20 per month for 1.7 billion tokens to $120 per month for 9.8 billion tokens.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.