Qwen 2.5 VL 72B Instruct: benchmark results
Alibaba Qwen 2.5 VL 72B vision-language instruction-tuned checkpoint. Provider: Alibaba. Released 2025-01-26. Access: Open.
Aggregate Rating 1621 ± 4, rank #383 of 1635 rated models, from 905 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Drive-P2D - Key Object (Object-1) | 68.06 | Accuracy (%; identify the object in the scene that most infl | 100 |
| EgoPro-Bench - Object | 84.01 | F1 (%; alert timing for the appearance of target objects, 20 | 100 |
| EgoSAT - Short-Horizon Anticipation | 43.06 | Multiple-choice accuracy (%) on the next action a short time | 100 |
| EgoSAT - Short-Horizon Recall | 51.87 | Multiple-choice accuracy (%) on the action a short time ago; | 100 |
| Embodied3DBench - Grounding (2D) | 62.8 | Grounding (2D) task (object and part pointing and bounding b | 100 |
| GB/T-Bench - Diagnosis Recall - Normative References | 60.4 | Diagnosis recall (%; exact section, dimension and error-type | 100 |
| GMP - Dynamic Rules (Shopping Platform Reviews) | 77.85 | F1 (x100) of the unsafe judgement under GMP Task B rule set | 100 |
| KnowHal - Relation (Negative) | 73.9 | Accuracy (%; relation-dimension negative (false-premise) que | 100 |
| M3-VQA | 32.55 | IoU accuracy (%) between the predicted and gold answer sets | 100 |
| MATEO (Image+Text, Instructions) | 61 | Consistent temporal-execution-order accuracy (%) on MATEO's | 100 |
| MATEO (Text-only, Baseline) | 14 | Consistent temporal-execution-order accuracy (%) on MATEO's | 100 |
| MATEO (Text-only, Instructions) | 53 | Consistent temporal-execution-order accuracy (%) on MATEO's | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-2-5-vl-72b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-05.