Unsloth Brings MiniMax-M2.7's 230B Coding Agent to a Single Mac

MiniMax's newest agentic coding model is now runnable on a single high-memory workstation. Unsloth has published GGUF quantizations of MiniMax-M2.7, a 230B-parameter mixture-of-experts model that was previously out of reach for anyone without a serious GPU cluster. The 4-bit dynamic quant fits in 108GB, so the model runs on a 128GB unified-memory Mac or a modest GPU-plus-RAM combo.

MiniMax-M2 packs 230 billion total parameters but activates only 10 billion at a time, keeping inference fast and cheap while preserving general intelligence. That tiny active footprint is what makes end-to-end tool use viable at this scale. M2.7 is the latest iteration in the series, and MiniMax pitches it as their first model deeply participating in its own evolution, capable of building complex agent harnesses and completing elaborate productivity tasks using Agent Teams, complex Skills, and dynamic tool search.

Benchmarks that actually matter here

M2.7 targets software engineering and office-style productivity rather than open-ended chat. On coding evals, it hits SWE-Pro at 56.22% and Terminal Bench 2 at 57.0%, with additional strong results on SWE Multilingual (76.5) and Multi SWE Bench (52.7). For agentic office work, it scores an ELO of 1495 on GDPval-AA, which MiniMax claims is the highest among open-source models, and maintains a 97% skill adherence rate while working with over 40 complex skills, each exceeding 2,000 tokens.