AbstentionBench - subjective - MoralChoice - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek R1 Distill Llama 70B | 100 |
| 2 | Llama-3.1-Tulu-3-70B | 98.54 |
| 3 | Llama-3.1-Tulu-3-70B-DPO | 98.51 |
| 4 | O1 | 98.51 |
| 5 | Llama 3.1 405B Instruct | 98.45 |
| 6 | GPT-4o | 98.34 |
| 7 | Llama 3.3 70B Instruct | 98.21 |
| 8 | Llama 3.1 70B Instruct | 98.21 |
| 9 | Llama-3.1-Tulu-3-8B-DPO | 97.93 |
| 10 | Llama-3.1-Tulu-3-8B | 97.81 |
| 11 | Qwen 2.5 32B Instruct | 97.8 |
| 12 | Llama-3.1-Tulu-3-8B-SFT | 97.77 |
| 13 | Llama 3.1 8B Instruct | 97.73 |
| 14 | Gemini 1.5 Pro | 97.63 |
| 15 | Mistral 7B Instruct (v0.3) | 96.79 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-subjective-moralchoice-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.