Latest / Frontier models and safety
MiniMax releases M3, a 1-million-token multimodal model for coding agents
See how this story connectsMiniMax released M3, a native multimodal model with about 428 billion parameters (23 billion active) and a 1-million-token context, five times its predecessor’s. MiniMax said its sparse attention design gives 9 times faster prefill and 15 times faster decoding than M2 at 1 million tokens, cutting per-token compute to a twentieth.
Why it matters
Long-context coding agents got cheaper to run on MiniMax’s model, by the company’s own figures.
- When
- 1 Jun 2026
- ID
WTA-N-2026-0209- Where
- China
- Who
- MiniMax
Summary by wheresthe.ai, from Hugging Face and the reporting credited above. Checked against the source on 10 Oct 2026.
Source copy: download the page as saved on 18 Oct 2026 (PDF). Also held by the Internet Archive.
Cite this story
WTA-N-2026-0209: “MiniMax releases M3, a 1-million-token multimodal model for coding agents.” wheresthe.ai, 1 Jun 2026. https://wheresthe.ai/d/cn-minimax-m3-release-2026-06/ (source: Hugging Face: MiniMaxAI/MiniMax-M3).
Spotted an error? Email [email protected].