VARM-Bench - Target Extraction: leaderboard
Part of VARM-Bench (Yuan et al., 2026) on moderation rationales for Chinese social-media text, zero-shot on the 1,600-item test set: mean normalized character-level F1 in percent of the free-text abuse target anchored in the rationale. Unparseable outputs count as zero; higher is better.
Metric: Target character-F1 (%; mean normalized character-overlap F1 of the extracted free-text target); deduplicated 1,600-item test set of Chinese social-media posts and comments; the model writes one rationale with six anchored decisions (target, target type, target explicitness, author stance, harmfulness label, fine-grained category) that a deterministic parser scores, unparseable outputs counting as wrong; deterministic decoding; zero-shot prompt. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 69.1 | #49 |
| 2 | Qwen 3.7 Max | 61.7 | #145 |
| 3 | DeepSeek V4 Pro | 61.1 | #188 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=varm-bench-target-extraction · How It Works · Data refreshed daily, snapshot 2026-11-13.