MaMe-Re (Speech Text + Panel Description): leaderboard
The MaMe-Re manga meme reply task (Kohita and Yoshioka, 2026) with each of the 400 candidate panels given by its speech text plus a GPT-4.1 description of the image: the model picks the funniest panel as a reply to a Japanese social-media post. Score@1 is the mean share of five crowdworkers who rated the chosen panel funny, in percent over 250 posts; higher is better.
Metric: Score@1 (%; mean share of five crowdworkers who rated the chosen manga-panel reply funny, 250 contexts, panels given by their speech text and a GPT-4.1 visual description). Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 32.5 | #163 |
| 2 | GPT-5 Mini (2025-08-07) | 30 | #341 |
| 3 | Gemini 2.5 Pro | 29.4 | #262 |
| 4 | GPT-OSS-120B | 28.4 | #699 |
| 5 | Gemini 2.5 Flash | 27.6 | #452 |
| 6 | Claude Sonnet 4 (20250514) | 26.1 | #386 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mame-re-speech-text-plus-panel-description · How It Works · Data refreshed daily, snapshot 2026-11-12.