Llama 3 8B UltraMedical: benchmark results
Provider: Meta. Access: Open.
Unified ELO 1414 ± 29, rank #1164 of 1639 rated models, from 16 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 66.17 | Accuracy (%) (5-shot) | 86.1 |
| CureMed-Bench - Language Consistency | 47.03 | Language consistency (%; share of final answers written in t | 64.6 |
| Open LLM Leaderboard v1 - GSM8K | 44.81 | Accuracy (%) (5-shot) | 58.8 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 52.32 | MC2 (%) (0-shot) | 55.4 |
| Open LLM Leaderboard v1 - ARC Challenge | 61.52 | Normalized accuracy (%) (25-shot) | 53.2 |
| Open LLM Leaderboard v1 - HellaSwag | 82.37 | Normalized accuracy (%) (10-shot) | 50.8 |
| CureMed-Bench | 35.29 | Logical accuracy (%; share of final answers a GPT-4o judge m | 45.8 |
| Open LLM Leaderboard v1 - WinoGrande | 75.22 | Accuracy (%) (5-shot) | 40.9 |
| MedRoundsQA - Multi-Turn Diagnosis | 22.21 | Accuracy (%) | 28.6 |
| MedRoundsQA - Single-Turn Diagnosis | 50 | Accuracy (%) | 28.6 |
| EHRBench - Treatment | 43.09 | Accuracy (%) on the treatment-decision questions, EHR-ground | 3.3 |
| EHRBench - Diagnosis | 19.14 | Accuracy (%) on the diagnosis-decision questions (complete a | 0 |
Interactive version: theaggregate.ai/model?slug=llama-3-8b-ultramedical · How It Works · Data refreshed daily, snapshot 2026-10-09.