Hermes 4.3 (36B)
The community's neutrally-aligned open model — 36B parameters, 512K context, Apache 2.0, built for local inference.
About
Hermes 4.3 is Nous Research's flagship open-weight model series entry: a 36B dense model based on ByteDance's Seed-OSS-36B, post-trained with an extended context and hybrid reasoning mode. It was the first production model trained on Nous's Psyche decentralized training network.
Important distinction: "Hermes" is also the name of Nous's self-improving agent runtime (Hermes Agent, MIT, v0.21.5 as of Sept 2026) — the runtime is not the model. No Hermes 5 model release has been found as of Oct 13, 2026.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Nous-reported (niche, provider-reported, older-generation comparisons):
- RefusalBench: 74.6% answered (non-reasoning) — SOTA among non-abliterated models per Nous, vs GPT-5 (11.34%) and Opus 4.1 (15.38%)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Apache 2.0 open weights and a genuine niche (neutral alignment, local inference) with an active community. But it is a 10-month-old community-lab release from a small team: benchmarks are provider-reported on a self-created benchmark, there is no official API or pricing, and the Hermes name is easily confused with the agent runtime. Usable with reservations, especially for local/steerable deployments rather than frontier capability.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.4/10
- Maintenance · recency, releases, issue load
- 4.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · HF top match: bartowski/NousResearch_Hermes-4-14B-GGUF (community upload counts as proxy where official repo is gated)
- · HF downloads 68,709 (peer-normalized)
- · released 1016 days ago
- · manual review: from 2024; small team; benchmark is self-created