MiniMax Open Sourced A Coding Model Anyone Can Run
MiniMax M3 scores like last year's Claude for a sixteenth the cost, here's what's free, what it takes to run it yourself, what renting costs and how to plug it into Claude Code. MiniMax released M3, a 428-billion-parameter open-weight coding model free to download, and open-sourced MiniMax Code, a terminal coding agent that can use it. On Artificial Analysis's independent Intelligence Index, M3 scores 29, the same score Artificial Analysis estimates for Claude Opus 4.5 from November 2025, and far behind today's Opus 5.5 (58). Running the full benchmark cost $538 on M3 against $8,708 on Opus 5.5, about 16 times more. The license is free for personal use; commercial products need a "Built with MiniMax M3" credit and notice, or written permission above $20M yearly revenue. The model weights, all 428 billion parameters, with 23 billion activated at a time, sit on Hugging Face under the MiniMax Community License. Even Unsloth's smallest compressed copy (1-bit) is 128 GB and needs more than that in memory (Unsloth suggests at least 133 GB of RAM), which rules out ordinary laptops and desktops: running it at home takes a workstation or server built with that much memory. Renting M3 cost $0.30 per million input tokens and $1.20 per million output tokens on 27 September 2026 (MiniMax's price for requests up to 512K input tokens, labelled a permanent 50% discount), about 13 to 17 times cheaper than Claude Opus 5.5 ($4/$20 per million). MiniMax's official guide shows how to point Claude Code at MiniMax's address with a MiniMax key, either in Claude Code's settings file or with MiniMax's setup tool, so you keep the same terminal interface while your requests go to MiniMax instead of Anthropic. MiniMax Code (on GitHub; its first-party code is MIT-licensed) installs on an ordinary computer and works with a MiniMax account or a model you bring yourself. On Terminal-Bench 4.0, Artificial Analysis measured M3 solving 2% of the tasks against just under 60% (59.6%) for Opus 5.5; MiniMax's own 66% was on the older Terminal-Bench 2.1. This video is for builders weighing cost against capability and deciding whether to rent M3 inside Claude Code or stay with today's Claude. Chapters: 0:00 Intro 1:35 What's free 3:20 Hardware 5:00 Renting 6:41 Vs Claude 9:26 Conclusion Tools & resources mentioned: - MiniMax M3: https://huggingface.co/MiniMaxAI/MiniMax-M3 - MiniMax Code: https://github.com/MiniMax-AI/minimax-code - Unsloth MiniMax M3 GGUF: https://huggingface.co/unsloth/MiniMax-M3-GGUF - MiniMax's Claude Code guide: https://platform.minimax.io/docs/token-plan/claude-code - MiniMax Platform API: https://platform.minimax.io - Artificial Analysis: MiniMax-M3 vs Claude Opus 5.5: https://artificialanalysis.ai/models/comparisons/minimax-m3-vs-claude-opus-5-5 About The Stack The Stack helps you build with AI. Each video takes one tool, model, or workflow and shows how it works in a few focused minutes, with the real benchmarks and real costs. We go deep on Claude Code and Cursor for AI coding, AI agents and MCP servers, the open-source AI tools and GitHub repos most people miss, RAG and vector search, fine-tuning, and running local LLMs on your own machine with Ollama and LM Studio. We compare models like ChatGPT and Claude, test AI automation with Zapier, Make, and n8n, and flag the tools that actually ship. Subscribe for new breakdowns: https://www.youtube.com/@the-stack-ai?sub_confirmation=1 #minimax #Claude Code #AI models #open source #coding agents
