Hermes Agent Plugs Into NVIDIA NIM, Unlocks MiniMax M2.7

NousResearch's Hermes Agent v0.8.0 formalizes a product partnership with MiniMax, bringing M2.7 natively into its agent framework. M2.7 is a 230B sparse Mixture-of-Experts model that activates only 10B parameters per token — keeping inference fast and affordable at $0.30/M input tokens.

The model hits 78% on SWE-bench Verified, putting it in serious contention for coding and agentic workloads. It's hosted directly on NVIDIA NIM, so you can hit it through build.nvidia.com without managing your own deployment. NVIDIA's free tier covers 100+ models under rate limits — no credit card required. Grab a key at build.nvidia.com/settings/api-keys.

The Hermes + MiniMax pairing is confirmed live through the Nous Portal, making it a one-config setup for developers already in the Hermes ecosystem.

Why it matters: A competitive 230B MoE model, open-source agent framework, and a zero-cost entry point is a rare combination — and it just got easier to wire together.