Incompressible Knowledge Probes (IKP): leaderboard
Closed-book factual recall across seven obscurity tiers, from common facts to obscure Wikidata entities and researcher knowledge. Reports the unweighted mean of tier accuracies on all 1,311 cleaned v2 questions, with no hallucination penalty. Measures recalled knowledge, not a direct parameter count.
Metric: Mean tier accuracy (%). Source: 01.me. Saturation forecast: Around February 2027. 133 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) (Medium) | 86.19 | #90 (Gemini 3 Flash (Preview)) |
| 2 | Gemini 3.1 Pro (Preview) | 85.6 | #64 |
| 3 | Gemini 3 Flash (Preview) | 85.06 | #90 |
| 4 | GPT-5.5 | 82.56 | #33 |
| 5 | GPT-5.5 (Medium) | 82.56 | #33 (GPT-5.5) |
| 6 | Claude Fable 5 | 80.49 | #20 |
| 7 | Gemini 2.5 Pro (Preview) | 79.48 | #136 |
| 8 | Gemini 2.5 Pro (Medium) | 78.87 | #159 (Gemini 2.5 Pro) |
| 9 | GPT-5.4 Pro (xHigh) | 77.84 | #38 (GPT-5.4 Pro) |
| 10 | GPT-4.1 | 77.18 | #258 |
| 11 | GPT-5 Pro | 76.98 | #89 |
| 12 | O3 | 76.91 | #132 |
| 13 | Gemini 3.1 Flash Lite (Preview) | 76.8 | #201 |
| 14 | GPT-5 (Medium) | 76.52 | #105 (GPT-5) |
| 15 | GPT-5 | 75.82 | #105 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=incompressible-knowledge-probes-ikp · How It Works · Data refreshed daily, snapshot 2026-11-05.