NovGauge - Pairwise Novelty - Problem F1: leaderboard
Raw F1 for judging paper similarity on the problem dimension (technical challenges) in NovGauge pairwise judgments on paper pairs drawn from ICLR reviewer overlap claims and survey co-citations, with full-paper content (NovGauge, arXiv 2609.11234v1, Table 3). Percent, mean of three runs at the model's maximum reasoning setting; higher is better. Does not check evidence faithfulness; see Verified F1.
Metric: F1 (%). Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Flash Lite | 89.2 | #200 |
| 2 | Gemini 3.1 Pro (Preview) | 77.1 | #64 |
| 3 | Qwen 3.5 27B | 76.7 | #192 |
| 4 | Kimi K2.6 | 76.5 | #113 |
| 5 | Qwen 3.6 Plus | 74.9 | #124 |
| 6 | MiniMax-M2.7 | 73.3 | #270 |
| 7 | GPT-5.5 | 69.4 | #33 |
| 8 | GLM-5.1 | 69.3 | #109 |
| 9 | Claude Opus 4.7 | 66.2 | #51 |
| 10 | Gemma 4 31B (IT) | 66.2 | #178 |
| 11 | Claude Sonnet 4.6 | 66.1 | #94 |
| 12 | DeepSeek V4 Pro | 60.3 | #112 |
| 13 | DeepSeek V4 Flash | 60 | #155 |
| 14 | GPT-5.4 Mini | 56.7 | #228 |
| 15 | Qwen 3 32B | 56.6 | #449 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=novgauge-pairwise-novelty-problem-f1 · How It Works · Data refreshed daily, snapshot 2026-11-05.