Long-horizon game control on StarDojo Lite-100
20.3Lite-100 Success Rate (SR)SPIKE
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SPIKELLM Backend=GPT-5.42026.05 | 20.3 | 23.7 | 107.8 | 27.9 | 3.5 | |
| SPIKELLM Backend=Default Qwen3.5-397B2026.05 | 18 | 21.6 | 168.1 | 43.5 | 2.75 | |
| SPIKELLM Backend=Gemini-3.1-pro2026.05 | 17.7 | 20.3 | 134.3 | 33.8 | 2.66 | |
| CRADLELLM Backend=Qwen3.5-397B-A17B2026.05 | 13 | 10.7 | 372.5 | 73.5 | 2.49 | |
| StarDojoLLM Backend=Qwen3.5-397B-A17B2026.05 | 12.3 | 12.3 | 295.9 | 48.7 | 2.37 | |
| Reflexion-likeLLM Backend=Qwen3.5-397B-A17B2026.05 | 8.7 | 10.4 | 178.5 | 35.9 | 2.02 | |
| Voyager-likeLLM Backend=Qwen3.5-397B-A17B2026.05 | 7.3 | 8.1 | 238.3 | 56.5 | 1.73 | |
| ReAct-likeLLM Backend=Qwen3.5-397B-A17B2026.05 | 6.7 | 9.6 | 102.6 | 24.5 | 1.61 |