Atlas

Models

← All models

Hermes 3 Llama 3.1 70B

Nous Research · 2024-07-29 · 70.6B parameters

Hermes 3 Llama 3.1 70B is Nous Research's instruction-tuned fine-tune of Meta's Llama 3.1 70B, released in July 2024. It was fine-tuned on an approximately 390-million-token instruction dataset with primarily synthetically generated responses, retains the base model's 131,072-token context window, and adds agentic function calling and structured outputs. Because the dataset size is approximate and the base model's cumulative pretraining exposure is not exact, no exact training-token count is recorded.

Benchmark scores

BenchmarkScore
Artificial Analysis Intelligence Index5.1
BBH (Open LLM Leaderboard v2)67.6
Capability119.7
GPQA (Open LLM Leaderboard v2)36.2
GPQA Diamond40.1
Humanity's Last Exam (Text-Only)4.1
IFEval (Open LLM Leaderboard v2)76.6
LiveCodeBench (Artificial Analysis)18.8
MATH Level 5 (Open LLM Leaderboard v2)21.0
MATH-500 (Artificial Analysis source)53.8
MMLU Pro57.1
MMLU-Pro (Open LLM Leaderboard v2)47.3
MuSR (Open LLM Leaderboard v2)49.5
SciCode (Artificial Analysis)23.1
Loading Atlas data…