SealedBench: leaderboard
SealedBench is a private evaluation that is continuously rotated, and it keeps its tests, scoring weights and answer keys unpublished to resist training aimed at the benchmark. Models with a configurable thinking setting are run at its highest level. The published number is a single index blending capability with token efficiency, speed and cost, charted from 0 to 100, where higher is better.
Metric: Capability Index. Source: sealedbench.com. Saturation forecast: Around October 2027. 36 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 75 | #6 |
| 2 | Claude Fable 5.1 | 72 | #9 |
| 3 | Claude Fable 5 | 71 | #16 |
| 4 | Grok 4.6 | 70 | #25 |
| 5 | GPT-5.6 Sol | 69 | #18 |
| 6 | Claude Opus 5 | 69 | #12 |
| 7 | DeepSeek V4 Flash | 68 | #139 |
| 8 | Kimi K3 | 68 | #30 |
| 9 | Claude Opus 4.8 | 67 | #35 |
| 10 | GPT-5.6 Terra | 66 | #40 |
| 11 | Claude Opus 4.6 | 65 | #63 |
| 12 | Grok 4.5 | 64 | #49 |
| 13 | GPT-5.6 Luna | 63 | #72 |
| 14 | GLM-5.3 | 63 | #36 |
| 15 | GPT-5.5 | 62 | #29 |
No result here: #4 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=sealedbench · How It Works · Data refreshed daily, snapshot 2026-10-23.