MiniMax

MiniMax is the Shanghai-based AI lab known for shipping open-weight models with aggressive long-context engineering. Its June 1, 2026 release, MiniMax M3, is built on MiniMax Sparse Attention (MSA), which replaces full attention with KV-block selection and cuts per-token compute at 1M context to roughly one twentieth of the previous generation. M3 takes text, image, and video input across a 1,048,576 token context window, reports 59% on SWE-Bench Pro and 83.5 on BrowseComp, and is priced at $0.30 per million input and $1.20 per million output, around 5 to 10 percent of the cost of proprietary flagships. The headline benchmark runs used MiniMax's own infrastructure and agent scaffolding, so independent verification is pending, and the open weights are due on Hugging Face within about ten days of launch.