Counting Task on Spatial and Temporal Simulation Tasks Step-level
4.237Subtask AccuracyCortex (full harness)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Cortex (full harness)Level=Step-level2026.07 | 4.237 | 4.504 | 8.741 | |
| Cortex (harness on skills)Level=Step-level2026.07 | 3.527 | 4.348 | 7.875 | |
| Cortex (baseline)Level=Step-level2026.07 | 3.503 | 4.173 | 7.676 | |
| Cortex (w/o harness)Level=Step-level2026.07 | 3.435 | 4.213 | 7.648 | |
| GeminiLevel=Step-level2026.07 | 3.135 | 4.025 | 7.159 | |
| Qwen3-VL-8B-InstructLevel=Step-level2026.07 | 2.948 | 4.07 | 7.018 | |
| GPT-5Level=Step-level2026.07 | 2.79 | 3.429 | 6.22 |