BenchTable (dubesor): leaderboard
One person's 83+ hand-crafted tasks tested across 312+ models. Tracks pass/refine/fail rates, STEM ability, reasoning, and censorship. Honest personal testing - not industry-standard, but insightful.
Metric: Total Score (%). Source: dubesor.de. 347 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 86.1 | #90 |
| 2 | Gemini 3 Pro (Preview) | 85.3 | #66 |
| 3 | Claude Opus 4.7 | 83.3 | #49 |
| 4 | Claude Opus 4 (Thinking) | 82.8 | #163 (Claude Opus 4) |
| 5 | Claude Opus 4.1 (Thinking) | 82.7 | #140 (Claude Opus 4.1) |
| 6 | Gemini 3.1 Pro (Preview) | 82.6 | #58 |
| 7 | Claude Opus 4.1 | 81.9 | #140 |
| 8 | Claude Opus 4 | 81.5 | #163 |
| 9 | Claude Opus 4.6 | 81.2 | #63 |
| 10 | Claude Opus 4.5 | 81.2 | #83 |
| 11 | Claude Sonnet 4 | 79.9 | #199 |
| 12 | GPT-4 Turbo | 79.5 | #404 |
| 13 | Grok 4 | 79.3 | #181 |
| 14 | Gemini 2.5 Pro | 78.5 | #152 |
| 15 | GPT-4.5 (Preview) | 78.5 | #244 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=benchtable-dubesor · How It Works · Data refreshed daily, snapshot 2026-10-20.