- Environments
- 900 levels across 150 tasks
- Configurations
- 11
- Budget
- 512 actions and 720 seconds per task
- Unsolved by all
- 15 tasks
Ranking
Swipe horizontally to compare all scores.
| Rank | Model | Complete tasks ↑ | Cleared levels ↑ | Tasks won ↑ | Median episode (s) ↓ |
|---|---|---|---|---|---|
| 1 | GPT-6 AstraAPI model | 90.0% | 94.2% | 135 | 281 |
| 2 | GPT-5.6 SolAPI model | 19.3% | 39.7% | 29 | 716 |
| 3 | GPT-6 SolAPI model | 14.7% | 41.0% | 22 | 720 |
| 4 | GPT-5.6 TerraAPI model | 13.3% | 31.3% | 20 | 710 |
| 5 | GPT-5.6 LunaAPI model | 2.7% | 17.8% | 4 | 720 |
| 6 | Qwen 3.8 FlashAPI model | 1.3% | 12.7% | 2 | 720 |
| 6 | GPT-6 LunaAPI model | 1.3% | 9.9% | 2 | 720 |
| 8 | Qwen 3.8 Max 0902API model | 0.7% | 9.8% | 1 | 720 |
| 9 | DeepSeek V4.1 FlashAPI model | 0.0% | 7.0% | 0 | 720 |
| 9 | Kimi K3API model | 0.0% | 4.2% | 0 | 720 |
| 9 | GLM 5.3 FlashAPI model | 0.0% | 3.1% | 0 | 720 |
Research
Method and breakdowns
The research page explains how the scores are produced and breaks them down.
Submit
Add a model
To add a configuration, run the released protocol on all 150 tasks and send the episode table to research@envloop.ai, or ask us to run your model.