Atomic Perception of Transitions on APT-BENCH (test)
53.4RecallAPT-Tune
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| APT-TuneBackbone=InternVL3.5-8B, Evaluation Protocol=+APT-Tune2026.06 | 53.4 | 71.2 | 19 | 40.4 | |
| APT-TuneBackbone=Qwen3-VL-8B, Evaluation Protocol=+APT-Tune2026.06 | 48.6 | 66.2 | 27 | 36.6 | |
| APT-TuneBackbone=InternVL3.5-4B, Evaluation Protocol=+APT-Tune2026.06 | 48 | 63.5 | 23 | 35 | |
| APT-TuneBackbone=InternVL3.5-2B, Evaluation Protocol=+APT-Tune2026.06 | 46.8 | 61.5 | 25 | 32.8 | |
| APT-TuneBackbone=Qwen3-VL-4B, Evaluation Protocol=+APT-Tune2026.06 | 42.5 | 57.8 | 33 | 31.5 | |
| APT-TuneBackbone=Qwen3-VL-2B, Evaluation Protocol=+APT-Tune2026.06 | 38.1 | 52.1 | 38 | 28.1 | |
| InternVL3.5-2BEvaluation Protocol=Zero-shot2026.06 | 14 | 16 | 67 | — | |
| InternVL3.5-4BEvaluation Protocol=Zero-shot2026.06 | 13 | 15 | 75 | — | |
| InternVL3.5-8BEvaluation Protocol=Zero-shot2026.06 | 13 | 15 | 80 | — | |
| Qwen3-VL-8BEvaluation Protocol=Zero-shot2026.06 | 12 | 14 | 90 | — | |
| Qwen3-VL-4BEvaluation Protocol=Zero-shot2026.06 | 11 | 13 | 95 | — | |
| Qwen3-VL-2BEvaluation Protocol=Zero-shot2026.06 | 10 | 12 | 100 | — |