GPT-6 Astra vs Hy4
GPT-6 Astra leads by 5.3 capability points (≈95% interval +2.7 to +7.9): a clear lead over Hy4 · P(higher) ≈ 100%
Wins 6 of 10 shared benchmarks (4 lost)
Each line runs through one model’s priced reasoning settings, cheapest to most expensive; grey points are other models’ priced settings in the same range. Results whose reasoning effort the source didn’t report are left out wherever a model has named effort levels. Capability and relative cost are the published estimates (relative cost is cost per task as a multiple of GPT-5 (high)); hover a setting, or use the arrow keys, to set it beside the other model’s same setting.
GPT-6 AstravsHy4capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev
Benchmark wins6 – 4 · 10 shared
GPT-6 AstravsHy4wins by categoryData Nov 15, 2026Data: Mercor, Neel Nanda, DAXBench and 5 othersVillage LLM Atlasatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Hy4 minus GPT-6 Astra.
| Terminal-Bench-Science 0.1 | Science | 68.1 | 19.8 | −48.3 |
| NoCoT-Bench NCRI 15.2 | General QA | 159 | 94.0 | −35.7 |
| Humanity's Last Exam Diamond | General QA | 60.6 | 32.7 | −27.9 |
| Terminal-Bench 4.0 | Code | 58.2 | 44.4 | −13.8 |
| NoCoT-Bench NCKI (kspine v3) | General QA | 131 | 111 | −13.7 |
| FoodTruck Bench | Games | 57761 | 67598 | +12.6 |
| APEX-Agents 1.1 | Professional Work | 75.1 | 85.1 | +10.0 |
| DAXBench v2 | Code | 96.8 | 87.2 | −9.6 |
| Harvey LAB (Mercor run, undisclosed ~213-task cohort) | Legal | 9.2 | 13.6 | +4.4 |
| τ³-Banking (Mercor run, bm25_grep retrieval) | Agents | 46.7 | 48.1 | +1.4 |
GPT-6 AstravsHy4normalised score marginsData Nov 15, 2026Data: Mercor, Neel Nanda, DAXBench and 5 othersVillage LLM Atlasatlas.village.dev