BoxComm (Punch Events) - Play-by-Play: leaderboard

BoxComm boxing commentary generation when the requested category is play-by-play (real-time descriptions of the ongoing action), with video plus detected punch events as text as input. Percent of generated sentences that GPT-5.2 judges consistent with the reference commentary, higher is better.

Metric: Content consistency accuracy (%) on the sentences whose requested category is play-by-play (real-time action description); input video plus the 16 most recent detected punch events as text (the authors EIC-Gen input); category-conditioned commentary generation on BoxComm-Eval (the last 40 of 445 World Boxing Championship broadcast matches, 5.6K professional commentary sentences): from the 4 seconds of video before the target moment, the 8 previous commentary sentences and a requested category (play-by-play, tactical or contextual) the model writes one commentary sentence, and GPT-5.2 judges whether it conveys the core action or tactical meaning of the reference sentence; zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Around May 2027. 3 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-4o Mini18.5#1667
2Qwen 3 VL 8B Instruct16.4#1029
3LLaVA-OV-7B (BoxComm checkpoint unspecified)13.3

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=boxcomm-punch-events-play-by-play · How It Works · Data refreshed daily, snapshot 2026-11-21.