← All comparisons
Tri-21B-think Preview vs Qwen3 4B (Reasoning)
Trillion Labs vs Alibaba — side-by-side benchmark comparison
| Tri-21B-think Preview | Qwen3 4B (Reasoning) | |
|---|---|---|
| Intelligence Index | 20.0 | 14.2 |
| Coding Index | 7.4 | — |
| Math Index | — | 22.3 |
| Output speed (tok/s) | 0.0 | 104.0 |
| Blended price ($/1M) | $0.00 | $0.40 |
| Time to first token (s) | 0.00s | 1.17s |
| aime | — | 65.7% |
| aime 25 | — | 22.3% |
| artificial analysis coding index | 7.40 | — |
| artificial analysis intelligence index | 20.00 | 14.20 |
| artificial analysis math index | — | 22.30 |
| gpqa | 53.8% | 52.2% |
| hle | 5.7% | 5.1% |
| ifbench | 47.1% | 32.5% |
| lcr | 14.7% | 0.0% |
| livecodebench | — | 46.5% |
| math 500 | — | 93.3% |
| mmlu pro | — | 69.6% |
| scicode | 17.8% | 3.5% |
| tau2 | 93.3% | 19.0% |
| terminalbench hard | 2.3% | — |
Benchmark data from Artificial Analysis.