← All comparisons
Hermes 4 - Llama-3.1 405B (Non-reasoning) vs DeepSeek V3.2 Exp (Reasoning)
Nous Research vs DeepSeek — side-by-side benchmark comparison
| Hermes 4 - Llama-3.1 405B (Non-reasoning) | DeepSeek V3.2 Exp (Reasoning) | |
|---|---|---|
| Intelligence Index | 17.6 | 32.9 |
| Coding Index | 18.1 | 33.3 |
| Math Index | 15.3 | 87.7 |
| Output speed (tok/s) | 40.8 | 0.0 |
| Blended price ($/1M) | $1.50 | $0.31 |
| Time to first token (s) | 0.73s | 0.00s |
| aime | — | — |
| aime 25 | 15.3% | 87.7% |
| artificial analysis coding index | 18.10 | 33.30 |
| artificial analysis intelligence index | 17.60 | 32.90 |
| artificial analysis math index | 15.30 | 87.70 |
| gpqa | 53.6% | 79.7% |
| hle | 4.2% | 13.8% |
| ifbench | 34.8% | 54.1% |
| lcr | 20.0% | 69.0% |
| livecodebench | 54.6% | 78.9% |
| math 500 | — | — |
| mmlu pro | 72.9% | 85.0% |
| scicode | 34.6% | 37.7% |
| tau2 | 26.6% | 33.9% |
| terminalbench hard | 9.8% | 31.1% |
Benchmark data from Artificial Analysis.