Nous Research drops Hermes 3 in sizes up to 405B
Nous Research just introduced Hermes 3, its latest open model series available in 8B, 70B, and 405B parameter sizes. Trained alongside Lambda, the upgrade targets sharper roleplaying, agentic workflows, reliable function calling, and long-context coherence. It offers developers a powerful, highly steerable generalist model right out of the box.
Nous Research just released Hermes 3, the latest iteration of its flagship generalist language model series.
Better agents and fewer guardrails
The new model arrives in three distinct sizes: 8B, 70B, and a massive 405B parameters. According to Nous Research, Hermes 3 delivers improvements across the board, with specific focus on roleplaying, agentic workflows, reliable function calling, multi-turn chats, and long-context coherence.
Trained in partnership with cloud provider Lambda, the team published a comprehensive technical report detailing its training run and capabilities. Lambda is currently hosting Hermes 3 for free in their chat web interface, and users can also test the 405B model inside the Nous Research Discord community.
Why it matters
Frontier open-source releases from mega-labs often come wrapped in cautious, suffocating guardrails that stifle creative roleplay and rigid instruction following. Hermes 3 takes the opposite stance by focusing on being directly aligned to the user.
Handing the open-source community a finely tuned 405B model with dependable tool use gives developers a real alternative to proprietary API lock-in. While self-hosting the 405B weight remains out of reach for individual setups, the 8B and 70B versions keep things accessible, and the free hosted chat makes testing it a no-brainer right now.
Sources (1)
- twitter.com(opens in a new tab)Originaltwitter.com