EsoLang-Bench: Evaluating Genuine Reasoning in LLMs via Esoteric Languages

Source: Hacker News

Motivation

Current benchmarks for large language model (LLM) code generation primarily evaluate mainstream languages like Python, where models benefit from massive pretraining corpora. This leads to inflated accuracy scores that may reflect data memorization rather than genuine reasoning ability.

EsoLang‑Bench

EsoLang‑Bench is a benchmark of 80 programming problems across five esoteric languages—Brainfuck, Befunge‑98, Whitespace, Unlambda, and Shakespeare—where training data is 5,000 to 100,000× scarcer than Python.

Evaluation

We evaluated five frontier models using five prompting strategies and two agentic coding systems.

Results

• The best‑performing model achieves only 3.8 % overall accuracy, compared to ~90 % on equivalent Python tasks.
• All models score 0 % on problems above the Easy tier.
• Whitespace remains completely unsolved (0 % across all configurations).
• Self‑reflection provides essentially zero benefit.

These results reveal a dramatic gap between benchmark performance on mainstream languages and genuine programming ability, suggesting that current LLM code generation capabilities are far narrower than headline metrics imply.