Descriptions:
David Ondrej demonstrates how to run the Hermes Agent — an open-source AI agent framework with over 180,000 GitHub stars — at dramatically lower cost by pairing it with Minimax M3, a recently released model the video positions as the first to exceed the traditional price-to-capability curve. On SWE-Pro, Minimax M3 reportedly outperforms both Gemini 3.1 Pro and GPT-5.5; on Browse Comp, which is directly relevant to Hermes’s deep-research mode, it matches GPT-5.5 while costing a fraction of the price.
The efficiency comes from Minimax Sparse Attention (MSA), an architecture that selectively reads only contextually relevant tokens rather than the full window on every pass — enabling a 1-million-token context at roughly 1/20th the standard compute cost. The pricing differential is significant: Minimax M3 runs at $0.60 per million input tokens and $2.40 per million output tokens versus Claude Opus 4’s $5/$25, with an additional 50% launch discount currently active. At the $20/month subscription tier, Minimax provides approximately 1.7 billion tokens per month.
The walkthrough covers full installation of Hermes Agent via a one-line macOS terminal command, connecting Minimax M3 as the backend provider in both Hermes and OpenCode, and running simultaneous live demos — including a Doodle Jump HTML game and SVG animation generation — while tracking costs in real time (approximately one cent for early coding tasks). Note: this video is sponsored by Minimax.
📺 Source: David Ondrej · Published June 08, 2026
🏷️ Format: Hands On Build
