Llama3-ChatQA-1.5-70B: benchmark results
NVIDIA's Llama3-ChatQA-1.5-70B, built on the Llama 3 base family with additional conversational question-answering data for retrieval-augmented generation and tabular or arithmetic tasks. NVIDIA converted the Megatron-trained checkpoint to Hugging Face format. Provider: NVIDIA. Access: Open.
Aggregate Rating 1496 ± 36, rank #2413 of 10916 rated models, from 7 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 77.91 | Accuracy (%) (5-shot) | 99 |
| Open LLM Leaderboard v1 - HellaSwag | 88.19 | Normalized accuracy (%) (10-shot) | 90.1 |
| Open LLM Leaderboard v1 - WinoGrande | 82.87 | Accuracy (%) (5-shot) | 86 |
| Open LLM Leaderboard v1 - GSM8K | 65.5 | Accuracy (%) (5-shot) | 80.2 |
| Open LLM Leaderboard v1 - ARC Challenge | 67.32 | Normalized accuracy (%) (25-shot) | 76.4 |
| ChineseSafe Benchmark | 65.29 | Accuracy (%) | 50 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 46.96 | MC2 (%) (0-shot) | 36 |
Interactive version: theaggregate.ai/model?slug=llama3-chatqa-1-5-70b · How It Works · Data refreshed daily, snapshot 2026-11-08.