MiniMax M3 Emerges as Leading Open‑Weight 1M‑Context AI Model
Published on: June 8, 2026
Researchers at MiniMax recently unveiled MiniMax M3, an open‑weight language model equipped with a one million token context window. Launched on June 1, 2026, this model marks a milestone in scaling long‑context capabilities for AI systems. It integrates text, image, and video inputs natively, positioning itself uniquely in the multimodal model landscape.
M3 shows strong performance on coding benchmarks, achieving a score of 59.0% on SWE‑Bench Pro. This result edges out GPT‑5.5’s 58.6% and Google’s Gemini 3.1 Pro at 54.2%, though it still trails behind Claude Opus 4.8, which scored 69.2%. These results underscore M3’s competitive strength despite being open‑weight.
A key innovation behind M3 is the MiniMax Sparse Attention (MSA) mechanism, which dramatically reduces per‑token compute. Compared to its predecessor M2, MSA cuts computational cost by about twenty‑fold at a one‑million token context. It also accelerates processing, offering nearly ten times faster prefill speed and over fifteen times faster decoding.
MiniMax has also made the model accessible via API, priced between approximately $0.60 to $2.40 per million input and output tokens. This cost is estimated to be just 8–20% of the price of leading proprietary U.S. models. The developers have committed to releasing the model weights and a technical report within ten days of the announcement, signaling a strong commitment to transparency and reproducibility.
MiniMax M3’s emergence highlights significant shifts in AI model development. By offering high-performance capabilities in an open‑weight framework, it supports broader experimentation and access. Its long‑context ability enables applications demanding extended memory or persistent reasoning, such as document analysis, long‑form content generation, or multimodal storytelling. The sparse attention innovation presents an efficient path forward for large‑context modeling.
Looking ahead, open‑weight models like M3 could accelerate innovation in fields constrained by access to proprietary systems. Researchers, startups, and open‑source communities stand to benefit from such benchmarks, potentially narrowing the gap between public and private AI capabilities.
No comments yet.