R-4B: benchmark results
Provider: Tencent. Released 2025-08-05. Access: Open.
Aggregate Rating 1620 ± 26, rank #828 of 10916 rated models, from 156 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| DatBench - Counting | 92.7 | Accuracy (%) | 100 |
| DatBench - Math & Logic | 43.4 | Accuracy (%) | 100 |
| OpenVLM MathVista - FQA | 81.8 | Accuracy (%) | 99 |
| OpenVLM MMBench V1.1 CN - Structuralized Imagetext Understanding | 88.1 | Accuracy (%) | 98.9 |
| OpenVLM MMStar - Logical Reasoning | 81.6 | Accuracy (%) | 98.9 |
| OpenVLM MMBench V1.1 CN - OCR | 100 | Accuracy (%) | 98.6 |
| OpenVLM MMStar - Fine-Grained Perception | 64.4 | Accuracy (%) | 98.6 |
| OpenVLM MathVista - MWP | 90.9 | Accuracy (%) | 98.5 |
| OpenVLM MathVista - STA | 88.7 | Accuracy (%) | 98.5 |
| OpenVLM MMMU - Art | 86.7 | Accuracy (%) | 98.2 |
| OpenVLM MMBench V1.1 CN - Attribute Comparison | 90.7 | Accuracy (%) | 97.5 |
| OpenVLM MMBench V1.1 EN - Structuralized Imagetext Understanding | 87.2 | Accuracy (%) | 97.5 |
Interactive version: theaggregate.ai/model?slug=r-4b · How It Works · Data refreshed daily, snapshot 2026-11-08.