NovGauge - Pairwise Novelty - Problem Verified F1: leaderboard
Verified F1 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: true positives count only if the cited evidence exists verbatim and an LLM judge finds it logically supports the stated reason (NovGauge, arXiv 2609.11234v1, Table 3). Percent; higher is better.
Metric: Verified F1 (%). Source: arxiv.org. Saturation forecast: Around December 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 61 | #33 |
| 2 | GPT-5.4 Mini | 37.3 | #228 |
| 3 | Gemma 4 31B (IT) | 31.7 | #178 |
| 4 | Kimi K2.6 | 25.7 | #113 |
| 5 | GLM-5.1 | 25.3 | #109 |
| 6 | Gemini 3.1 Pro (Preview) | 24.7 | #64 |
| 7 | Gemini 3.1 Flash Lite | 22.9 | #200 |
| 8 | Qwen 3 32B | 16.2 | #449 |
| 9 | Claude Sonnet 4.6 | 13.4 | #94 |
| 10 | Qwen 3.6 Plus | 12.3 | #124 |
| 11 | DeepSeek V4 Pro | 11.4 | #112 |
| 12 | Qwen 3.5 27B | 10.6 | #192 |
| 13 | Claude Opus 4.7 | 10.5 | #51 |
| 14 | MiniMax-M2.7 | 4.4 | #270 |
| 15 | Qwen 3 8B | 3.9 | #678 |
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-verified-f1 · How It Works · Data refreshed daily, snapshot 2026-11-05.