MiniMaxAI/MiniMax-M2.7 · Hugging Face

Read https://huggingface.co/MiniMaxAI/MiniMax-M2.7 for the benchmark figures on models/minimax-m2-7. As recorded by the reader: Only published score matching the allowed benchmark list is 'SWE-Pro 56.22%' (MiniMax-M2.7, 229B total / 10B active), corroborated by the HF page's auto eval result for ScaleAI/SWE-bench_Pro (56.2). Card/blog also report Terminal Bench 2 57.0%, SWE Multilingual 76.5, Multi-SWE-Bench 52.7, VIBE-Pro 55.6, NL2Repo 39.8, GDPval-AA ELO 1495, Toolathon 46.3, MM Claw 62.7, MLE Bench Lite 66.6% medal rate — omitted because they are not in the allowed name list or are a different version (source says 'Terminal Bench 2', not Terminal-Bench 2.1). No GPQA Diamond, AIME 2025, MMLU-Pro or LiveCodeBench scores are published for M2.7.