MiniMax M3, an open-weights multimodal model launched June 1, 2026, closes the performance gap with GPT-5.5 and Opus while offering aggressive pricing at $0.60 per million input tokens.

MiniMax M3 is a natively multimodal model (text, image, and video input, text output) with a 1 million token context window. It features MiniMax Sparse Attention, a mechanism that selectively processes key-value blocks to achieve roughly 9× speedup on prefill and 15× on decode at 1M tokens.

On coding tasks, M3 scores 59.0% on SWE-Bench Pro (behind Claude Opus 4.7 at 64.3% and ahead of Gemini 3.1 Pro at 54.2%), and in code audits matched GPT-5.5's rigor without filler. On abstract reasoning, the model scores in low single digits on ARC-AGI-2, consistent with the broader Chinese model family.

Pricing is $0.60 per million input tokens and $2.40 per million output tokens, with a 50% launch promo for the first week. The model is available through the MiniMax API, OpenRouter, and launch partners. Parameter count remains undisclosed, and weights were promised within 10 days of launch.