Long-horizon Task on Spatial and Temporal Simulation Tasks Episode-level
3.576Subtask AccuracyCortex (full harness)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Cortex (full harness)Level=Episode-level2026.07 | 3.576 | 3.804 | 7.38 | |
| Cortex (w/o harness)Level=Episode-level2026.07 | 3.004 | 3.922 | 6.925 | |
| GeminiLevel=Episode-level2026.07 | 2.774 | 3.87 | 6.644 | |
| Cortex (harness on skills)Level=Episode-level2026.07 | 2.743 | 4.123 | 6.866 | |
| GPT-5Level=Episode-level2026.07 | 2.711 | 4.285 | 6.996 | |
| Qwen3-VL-8B-InstructLevel=Episode-level2026.07 | 2.274 | 3.746 | 6.021 | |
| Cortex (baseline)Level=Episode-level2026.07 | 2.238 | 4.362 | 6.6 |