← All comparisons
DeepSeek V4 Flash (Reasoning, Max Effort) vs Llama 3.1 Tulu3 405B
DeepSeek vs Allen Institute for AI — side-by-side benchmark comparison
| DeepSeek V4 Flash (Reasoning, Max Effort) | Llama 3.1 Tulu3 405B | |
|---|---|---|
| Intelligence Index | 46.5 | 14.1 |
| Coding Index | 38.7 | — |
| Math Index | — | — |
| Output speed (tok/s) | 119.3 | 0.0 |
| Blended price ($/1M) | $0.17 | $0.00 |
| Time to first token (s) | 0.86s | 0.00s |
| aime | — | 13.3% |
| aime 25 | — | — |
| artificial analysis coding index | 38.70 | — |
| artificial analysis intelligence index | 46.50 | 14.10 |
| artificial analysis math index | — | — |
| gpqa | 89.4% | 51.6% |
| hle | 32.1% | 3.5% |
| ifbench | 79.2% | — |
| lcr | 63.0% | — |
| livecodebench | — | 29.1% |
| math 500 | — | 77.8% |
| mmlu pro | — | 71.6% |
| scicode | 44.9% | 30.2% |
| tau2 | 95.0% | — |
| terminalbench hard | 35.6% | — |
Benchmark data from Artificial Analysis.