HieroglyphBench: leaderboard
Photographs of hieroglyph columns from the Pyramid of Unas are shown to the model, which returns the signs it reads as a list of Gardiner sign list codes in reading order. The set holds 30 inscriptions, and each answer is compared with the ground truth sequence by Levenshtein distance in a single turn. The score is one minus the sign error rate, averaged equally over inscriptions, higher is better.
Metric: Sign accuracy (%). Source: boggs.tech. Saturation forecast: Around 2029. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 4 Argon | 69.7 | #8 |
| 2 | Gemini 3.7 Flash | 55 | #23 |
| 3 | Gemini 3.8 Flash | 54.6 | #21 |
| 4 | Gemini 3.5 Flash | 52.5 | #60 |
| 5 | Gemini 3 Flash (Preview) | 48.8 | #82 |
| 6 | GPT-6 Astra | 40 | #6 |
| 7 | Gemini 3.1 Pro (Preview) | 39.7 | #58 |
| 8 | Claude Opus 5.5 | 37.4 | #3 |
| 9 | GPT-6.1 Sol | 31.4 | #7 |
| 10 | Muse Wonder | 29.7 | #15 |
| 11 | GPT-5.6 Sol | 27.6 | #19 |
| 12 | GPT-6 Sol | 26 | #14 |
| 13 | Kimi K3 | 25.9 | #30 |
| 14 | Muse Spark 1.1 | 24.5 | #52 |
| 15 | Claude Fable 5 | 22.9 | #16 |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=hieroglyphbench · How It Works · Data refreshed daily, snapshot 2026-10-20.