Agnes 3.0 Flash vs Step 5 Preview
On the 10 benchmarks both have, Step 5 Preview scores higher on 10.
No benchmark publisher we read lists Agnes 3.0 Flash any more, so its figures stopped updating on 22 Sep 2026. Step 5 Preview's figures are current as of 13 Nov 2026. Each score shows the date it was published.
| Benchmark |  Agnes 3.0 Flash |  Step 5 Preview |
|---|---|---|
| AA Intelligence Index v4.3.2 | 35.5 11 Sep 2026 | 43.7 18 Sep 2026 |
| AA-LCR v1.1 | 81.0% 11 Sep 2026 | 88.3% 18 Sep 2026 |
| AA-Omniscience v1.1 | -10.6 11 Sep 2026 | 16.4 18 Sep 2026 |
| Humanity's Last Exam (AA) | 38.5% 11 Sep 2026 | 46.5% 18 Sep 2026 |
| SciCode (AA) | 51.6% 11 Sep 2026 | 58.9% 18 Sep 2026 |
| CritPt | 15.1% 19 Sep 2026 | 20.9% 18 Sep 2026 |
| GDPval-AA v2.1, normalised score | 52.5% 20 Sep 2026 | 54.3% 18 Sep 2026 |
| Terminal-Bench 4.0 (AA) | 7.1% 22 Sep 2026 | 33.3% 18 Sep 2026 |
| AA-Omniscience v1.1: accuracy | 25.5% 22 Sep 2026 | 36.4% 18 Sep 2026 |
| AA-Omniscience v1.1: non-hallucination | 51.6% 22 Sep 2026 | 68.7% 18 Sep 2026 |
| Blended price (per 1M tokens, 3:1) | $0.075 | $1.43 |
| Output speed | 317 tok/s | 135 tok/s |
See everything we last measured for Agnes 3.0 Flash, or Step 5 Preview's page for how it compares with models that are still measured. The full list is on the leaderboards.