Up to four models
Llama 3 vs Llama 2
On the index (GPT-4 = 100), Llama 3 scores 101 and Llama 2 95. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Llama 3 and Llama 2 are too close to call (6 points apart, within the uncertainty), ahead on 1 of 1 shared core tests.
Test by testthird-party results
Every result as a table
| Test | Llama 3 | Llama 2 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 40.6% | 26.3% |
| MMLU (HELM runs) | 79.3% | 69.5% |
| Maths | ||
| AIME-style maths (OTIS mock) | 4.3% | 0.0% |
| MATH (level 5) | 22.6% | 3.3% |
| Agents | ||
| DTBench | 54.2% | 41.6% |
Tests at least two of these models have taken. Bold rows are the core tests.