Qwen 2.5 VL 7B Instruct: benchmark results
Alibaba Qwen 2.5 VL 7B vision-language instruction-tuned checkpoint. Provider: Alibaba. Released 2025-01-26. Access: Open.
Aggregate Rating 1537 ± 3, rank #654 of 1656 rated models, from 1374 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentRx (Multimodal, Zero-Shot) - Length of Stay | 0.71 | AUROC (0-1; stay beyond seven days; MIMIC-IV ICU test split; | 100 |
| BanglaMedVQA (English) | 48.25 | Exact-match accuracy (%; mean of the Chest X-Ray and MedICaT | 100 |
| BanglaMedVQA (English) - Chest X-Ray | 46.5 | Exact-match accuracy (%; 1,000 questions on ChestX-ray14 rad | 100 |
| BanglaMedVQA (English) - MedICaT | 50 | Exact-match accuracy (%; 1,000 questions on MedICaT figures; | 100 |
| BanglaMedVQA (English, CoT) | 47 | Exact-match accuracy (%; mean of the Chest X-Ray and MedICaT | 100 |
| BanglaMedVQA (English, CoT) - Chest X-Ray | 48 | Exact-match accuracy (%; 1,000 questions on ChestX-ray14 rad | 100 |
| CogRail RailThreat (Cog-MRSI, Type-II Visual Prompt) | 42.81 | Overall F1 (%; harmonic mean of macro precision and macro re | 100 |
| CourtSI-Bench - Player-Zone | 58.5 | Exact-match accuracy (%) on the 82 player-to-court-zone rela | 100 |
| DatBench - Scene OCR | 82.6 | Accuracy (%) | 100 |
| ECAC | 45.25 | Teaching Toy Recognition Score (TTS, %): share of the teachi | 100 |
| ECAC - Caption Quality | 85.7 | Overall caption quality score (0-100) from a moonshot-v1-8k- | 100 |
| FGTR-Bench - Daytime Signal | 97.9 | Accuracy (%) on 750 daytime traffic-signal questions, 4-opti | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-2-5-vl-7b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.