SayNext-Bench (PC2K) - ROUGE-L: leaderboard
Metric: Mean ROUGE-L F1 (%, 0-100) of the predicted against the transcribed real next utterance; next-utterance anticipation: given the interviewer's question and up to 16 video frames of the speaker's reaction, the model writes the speaker's next answer, zero-shot with one shared prompt (temperature 0.7), on the 675-turn test split of SayNext-PC2K (post-match tennis press conferences; subject-dependent protocol); higher is better. Source: arxiv.org. Saturation forecast: Around 2032. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 14.62 | #333 |
| 2 | InternVL2-8B | 13.94 | #826 |
| 3 | Gemini 2.5 Flash | 11.95 | #237 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=saynext-bench-pc2k-rouge-l · How It Works · Data refreshed daily, snapshot 2026-10-11.