AbstentionBench - underspecified context - GPQA-Diamond - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 | 87.32 |
| 2 | GPT-4o | 85.71 |
| 3 | Llama 3.1 405B Instruct | 78.79 |
| 4 | Qwen 2.5 32B Instruct | 78.79 |
| 5 | Llama-3.1-Tulu-3-70B | 77.61 |
| 6 | Llama-3.1-Tulu-3-70B-DPO | 77.61 |
| 7 | O1 (Low) | 76.92 |
| 8 | O1 (High) | 72.73 |
| 9 | Llama 3.1 8B Instruct | 68.85 |
| 10 | Gemini 1.5 Pro | 65.57 |
| 11 | Mistral 7B Instruct (v0.3) | 64.62 |
| 12 | Llama 3.1 70B Instruct | 61.02 |
| 13 | Llama 3.3 70B Instruct | 53.57 |
| 14 | Llama-3.1-Tulu-3-8B-SFT | 44.83 |
| 15 | DeepSeek R1 Distill Llama 70B | 38.46 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-gpqa-diamond-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.