HumanAgencyBench - Correct Misinformation: leaderboard
HumanAgencyBench dimension Correct Misinformation: 500 prompts with a false or misleading claim mixed into the request, testing whether the assistant identifies and corrects it. o3-judged 0-10 rubric score averaged per model and shown on a 0-100 scale; higher is better.
Metric: HAB dimension score (0-100, o3-judged rubric). Source: github.com. Saturation forecast: Around February 2027. 38 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 88.56 | #51 |
| 2 | Claude Opus 4.6 | 86.04 | #67 |
| 3 | GPT-5 (High) | 85.28 | #105 (GPT-5) |
| 4 | GPT-5.5 | 84.1 | #33 |
| 5 | Claude Opus 4.5 (20251101) | 82.32 | #81 |
| 6 | Claude Sonnet 4.6 | 79.72 | #94 |
| 7 | GPT-5.4 | 79.66 | #83 |
| 8 | Claude Opus 4.1 (20250805) | 79.18 | #139 |
| 9 | GPT-5 | 73.28 | #105 |
| 10 | Claude Sonnet 4 (20250514) | 70.64 | #223 |
| 11 | Gemini 3.1 Pro (Preview) | 67.18 | #64 |
| 12 | Grok 4 | 66.84 | #189 |
| 13 | Kimi K2.6 | 65.48 | #113 |
| 14 | GPT-5.4 Mini | 60.88 | #228 |
| 15 | O3 (2025-04-16) | 59.48 | #125 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=humanagencybench-correct-misinformation · How It Works · Data refreshed daily, snapshot 2026-11-05.