CLIP-CC-Bench: leaderboard
CLIP-CC-Bench (EvalMG workshop at SIGIR 2026) asks video-language models for paragraph-level descriptions of 200 90-second movie clips with expert-narrated references. The score is the harmonic mean of paragraph-level cosine similarity and sentence-level matching F1, averaged over five embedding-model judges and scaled to 0-100; higher is better.
Metric: Mean HM-CF similarity (0-100): harmonic mean of paragraph-level cosine similarity and sentence-level greedy-matching F1 between the generated description and the expert-written reference, averaged over the 199 scored clips of the 200-clip set of 90-second movie clips and over five MTEB embedding-model judges (GTE-Qwen2-7B, KaLM-Gemma3-12B, Llama-Embed-Nemotron-8B, NV-Embed-v2, Qwen3-Embedding-8B), printed 0-1 and scaled by 100; uniform prompt, greedy decoding. Source: arxiv.org. Saturation forecast: Around February 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL2-8B | 58 | #3025 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=clip-cc-bench · How It Works · Data refreshed daily, snapshot 2026-11-12.