M3R-Bench - Source Prediction: leaderboard

Metric: BERTScore F1 (0-100) of the predicted central source concept against the human annotation, predicted jointly with metaphor presence, target and sentiment. Source: arxiv.org. Saturation forecast: Around 2029. 8 models tracked.

Top models

#ModelScore
1Claude Sonnet 4.671.35
2GPT-4o71.23
3GPT-5.567.02
4Qwen 3 VL 32B Instruct66.53
5Qwen 3 VL 8B Instruct66.4
6Kimi K2.563.51

Interactive version: theaggregate.ai/benchmark?slug=m3r-bench-source-prediction · How It Works · Data refreshed daily, snapshot 2026-09-26.