Code Mala Tang
Jun 6, 2026 · Artificial Intelligence
MiniMax M3 Sets New Benchmarks: 1M Context, 59% SWE‑Bench, 9‑15× Faster Multimodal Model
MiniMax unveiled its open‑source M3 model, delivering 1 million‑token context, 59 % SWE‑Bench Pro accuracy that outperforms GPT‑5.5 and Gemini 3.1 Pro, native multimodal desktop interaction, and a 9‑15× speed boost via MiniMax Sparse Attention, with pricing as low as $20 per month.
BenchmarkMSAMiniMax M3
0 likes · 11 min read