Autor: Jakub Rusinowski · Ostatnia aktualizacja: 10 lipca 2026
PREVIEW (June 2026, specs unverified). MiniMax's M3 — billed as the first open-weight model to combine frontier coding, a 1M-token context window, and native multimodal input, reportedly topping the open-weight SWE-Bench Pro board (~59%). A ~230B MoE with ~10B active per token, released under a modified MIT license. Self-hosting needs multi-GPU/Mac-Studio-class memory; most access is via API. Specs are single-source pending verification — see the Hugging Face org page.
| MiniMax M3 230B-A10B | Min 140 GB VRAM · Q4_K_M · 1,000,000 ctx · |
| MiniMax M3-VL | Min 140 GB VRAM · Q4_K_M · 1,000,000 ctx · |
The cheapest GPU that runs MiniMax M3 locally (min 140 GB VRAM) is the Apple M2 Ultra (192 GB).
Install Ollama then run: ollama run
Minimum VRAM: 140 GB. For best results use Q4_K_M quantization.
MiniMax M3 needs about 140 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: MiniMax M3 230B-A10B (140 GB, Q4_K_M); MiniMax M3-VL (140 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
MiniMax M3's smallest variant needs about 140 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
Q4_K_M is the best balance of quality and VRAM for MiniMax M3 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
Install Ollama, then run: ollama run . This downloads MiniMax M3 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.