← All comparisons
Hermes 4 - Llama-3.1 70B (Non-reasoning) vs Llama 3.2 Instruct 3B
Nous Research vs Meta — side-by-side benchmark comparison
| Hermes 4 - Llama-3.1 70B (Non-reasoning) | Llama 3.2 Instruct 3B | |
|---|---|---|
| Intelligence Index | 12.6 | 9.7 |
| Coding Index | 9.2 | — |
| Math Index | 11.3 | 3.3 |
| Output speed (tok/s) | 94.3 | 52.7 |
| Blended price ($/1M) | $0.20 | $0.15 |
| Time to first token (s) | 0.61s | 0.65s |
| aime | — | 6.7% |
| aime 25 | 11.3% | 3.3% |
| artificial analysis coding index | 9.20 | — |
| artificial analysis intelligence index | 12.60 | 9.70 |
| artificial analysis math index | 11.30 | 3.30 |
| gpqa | 49.1% | 25.5% |
| hle | 3.6% | 5.2% |
| ifbench | 29.0% | 26.2% |
| lcr | 2.0% | 2.0% |
| livecodebench | 26.9% | 8.3% |
| math 500 | — | 48.9% |
| mmlu pro | 66.4% | 34.7% |
| scicode | 27.7% | 5.2% |
| tau2 | 21.6% | 21.1% |
| terminalbench hard | 0.0% | — |
Benchmark data from Artificial Analysis.