Artificial Analysis lists Hermes 4 Llama-3.1 70B
Artificial Analysis has added Hermes 4 - Llama-3.1 70B to its model tracker, so this 70.6-billion-parameter open-weights release now has a page for intelligence, cost and speed comparisons.
Benchmarking firm Artificial Analysis has published a release page for Hermes 4 - Llama-3.1 70B. The page says the release contains two models, and each has different intelligence, performance and pricing characteristics. It does not say how the two differ.
According to the listing, the release:
- takes text input and produces text output
- has a 128k-token context window
- has 70.6B total parameters
- is licensed under the Meta Llama 3 Community License Agreement
- has its weights on Hugging Face
Artificial Analysis compares open-weights models only with other open-weights models in the same size band. At 70.6B parameters, Hermes 4 70B falls in the "medium" band (40B–150B). Reasoning models are compared against both reasoning and non-reasoning models. Non-reasoning models are compared only with each other.
The page has sections for the firm's Intelligence Index, cost per task, time per task and capability indexes. These indexes draw on evaluations including AA-Omniscience v1.1, AA-Briefcase v1.1, HLE-Diamond, AutomationBench-AA, GDP.pdf, Terminal-Bench 4.0, FrontierCode-AA 2.0 and CritPt v2 (Hard Subset).
The captured page does not include actual scores, prices or speed figures for either model. It also does not say who developed the release, when it came out, or which API providers serve it. For now, that leaves no basis for judging how it ranks against other medium-sized open-weights models.