ATLAS-Finance v1.1 Rubric Score
Professional Work · 2026-10-15
Partial-completion score on ATLAS-Finance v1.1.0: the weighted share of expert rubric criteria a deliverable meets after section gates and penalties, averaged over the 100 tasks and three rollouts per task. It credits substantial but imperfect work that the strict pass rate scores as failure, and is a component of the v1.1 factor group.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 71.7 |
| Claude Fable 5.1 | 71.5 |
| GPT-6 Astra | 70.0 |
| Kimi K3 | 64.4 |
| Grok 4.6 | 64.3 |
| GLM-5.3 | 61.9 |
| Gemini 3.8 Flash | 61.8 |
| GPT-5.6 Sol | 60.0 |
| Gemini 3.7 Flash | 58.2 |
| Gemini 3.1 Pro Preview | 34.8 |