TADDLE Benchmark: leaderboard
TADDLE Benchmark evaluates models on text-only expert-annotated peer reviews, asking them to detect deficient LLM-generated reviews at the binary and multi-label defect-type level. The pinned metric is F1 (points), measuring classification performance on the benchmark’s defect labels, based on the source-reported results table. Source: BenchmarkList’s attributed result table.
Metric: F1 (points). Source: benchmarklist.com. Saturation forecast: Around May 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-4.7 | 0.67 | #202 |
| 2 | Qwen 3 30B A3B | 0.59 | #496 |
| 3 | Qwen 3.5 Plus | 0.57 | #133 |
| 4 | DeepSeek V3.2 | 0.39 | #210 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=taddle-benchmark · How It Works · Data refreshed daily, snapshot 2026-11-05.