Hieraticbench
Hieraticbench - benchmark for AI reading ancient hieratic script
Last verified:
What is Hieraticbench?
HieraticBench is a public benchmark that tests whether leading AI models can identify and translate a sample of hieratic, the everyday handwriting of ancient Egypt, written by an Oxford Egyptologist. It presents a leaderboard-style record of failed attempts from 13 major models (Claude, GPT, Gemini, Grok, Kimi, Llama, Mistral, Qwen) and invites users to help improve AI hieratic literacy. It was created by Aly Moursy, founder of Veeza AI.
Hieraticbench pricing
Pricing model: Freemium
Hieraticbench pros
- Covers a wide range of models (13 including Claude, GPT, Gemini, Grok, Llama, Mistral, Qwen)
- Tests both consumer apps and APIs using a consistent methodology (same image, same plain question, multiple tries)
- Includes an API-based evaluation harness (HieraticBench) for reproducible testing
- Highlights a genuine, well-documented blind spot in multimodal LLMs with real expert-written source material
Hieraticbench cons
- Very narrow evaluation scope — a single hieratic sentence rather than a broad benchmark suite
- No visible pricing, documentation, or public dataset/code beyond the page itself
- Not a general-purpose tool; it's primarily a demonstration and call to action
Frequently asked questions about Hieraticbench
What is HieraticBench?
A benchmark testing whether AI models can identify and translate a hieratic sentence written by an Oxford Egyptology professor in 2022.
Did any AI model read the hieratic sentence correctly?
No. Across 13 tested models, none could correctly name the script, let alone translate it.
Who created HieraticBench?
Aly Moursy, founder of Veeza AI (YC F26), using hieratic text written by an Oxford Egyptologist.
How were the models tested?
The same image and plain question were asked in the consumer apps and via the API with multiple attempts each.