Pangram AI-Text Detection Hard Set (8-shot): leaderboard
Pangram hard-set AI-text detection — accuracy of models at distinguishing human-written from AI-generated text on adversarially hard samples, with 8 in-context examples.
Metric: Accuracy (%). Source: www.pangram.com. Saturation forecast: Around January 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 96.2 | #31 |
| 2 | Claude Sonnet 4.6 | 93.8 | #90 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=pangram-ai-text-detection-hard-set-8-shot · How It Works · Data refreshed daily, snapshot 2026-10-20.