C3-Bench - Relevance: leaderboard
C3-Bench (2026) evaluates context-aware change captioning: describing what changed between two images according to criteria that differ across 51 real-world change contexts. This board is relevance (0 to 10), whether the caption covers the changes the context asks for and ignores the rest, scored by a GPT-5.2 judge over 4,996 human-labelled image pairs and averaged over three judge runs. Higher is better.
Metric: Relevance (the caption covers the changes the context asks for and ignores the rest) (0-10) of change captions on the 4,996 human-labelled C3-Bench image pairs (51 change contexts across natural scenes, remote sensing, image editing and anomalies), scored by a GPT-5.2 judge against the reference description with the context-specific change criteria; mean of three judge runs; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 5.98 | #176 |
| 2 | GPT-5.1 | 5.98 | #229 |
| 3 | Gemini 3 Pro | 5.95 | #137 |
| 4 | Gemini 3 Flash | 5.86 | #164 |
| 5 | Gemini 2.5 Pro | 5.85 | #266 |
| 6 | GPT-4o | 5.83 | #741 |
| 7 | Gemini 2.5 Flash | 5.69 | #453 |
| 8 | InternVL3-78B | 5.59 | #738 |
| 9 | Qwen 3 VL 32B Instruct | 5.58 | #585 |
| 10 | Qwen 3 VL 8B Instruct | 5.54 | #981 |
| 11 | InternVL3-38B | 5.46 | #903 |
| 12 | GPT-4o Mini | 5.23 | #1623 |
| 13 | Llama 4 Scout Instruct | 5.19 | #1543 |
| 14 | Qwen 2.5 VL 7B Instruct | 5.07 | #1833 |
| 15 | Qwen 2.5 VL 72B Instruct | 5.06 | #846 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=c3-bench-relevance · How It Works · Data refreshed daily, snapshot 2026-11-13.