Level Progress on Super Mario Land (train)
1,644.43Performance (1,1)Odysseus
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| OdysseusBase Model=Qwen3-VL-8B-Instruct, Training Protocol=RL on SFT2026.05 | 1,644.43 | 1,430.88 | 1,603.36 | 1,352.3 | 1,528.51 | 1,511.9 | |
| Odysseus-ZeroBase Model=Qwen3-VL-8B-Instruct, Training Protocol=RL on base2026.05 | 1,545.5 | 1,222.69 | 1,551.57 | 1,262.18 | 1,192.71 | 1,354.93 | |
| GLM-4.6V (106B-A12B)Architecture=MoE, Parameters=106B-A12B2026.05 | 731.28 | 364.85 | 534.04 | 478.46 | 455.94 | 512.91 | |
| Claude-Sonnet-4.6Type=Proprietary2026.05 | 608.12 | 132 | 502.62 | 206.5 | 291.75 | 348.19 | |
| Gemini-3-FlashType=Proprietary2026.05 | 529.12 | 255.88 | 493.43 | 187.75 | 239.5 | 341.14 | |
| Qwen3-VL-8B-Instruct (base)Parameters=8B, Training Protocol=Instruction-tuned base2026.05 | 513.57 | 129.14 | 274.2 | 238.92 | 195.33 | 270.23 | |
| Odysseus-SFTBase Model=Qwen3-VL-8B-Instruct, Training Protocol=Supervised Fine-Tuning2026.05 | 479.47 | 90.92 | 300.76 | 245.01 | 192.69 | 261.77 | |
| InternVL3.5-241B-A28BArchitecture=MoE, Parameters=241B-A28B2026.05 | 442.41 | 78.12 | 390.11 | 188.61 | 196.4 | 259.13 | |
| Qwen3-VL-235B-A22B-InstructArchitecture=MoE, Parameters=235B-A22B2026.05 | 424.42 | 51.69 | 511.48 | 186.88 | 222.77 | 279.45 | |
| GPT-5.4Type=Proprietary2026.05 | 403.62 | 67.62 | 654.86 | 252.75 | 173.5 | 310.47 |