Lambda Calculus Benchmark for AI | Hacker News

Hacker News users debated a lambda-calculus benchmark for LLMs, saying the results are hard to trust without full testing details.

Commenters said open, local, and quantized model runs can vary a lot, making benchmark comparisons easy to misread.

Several users argued that small open models are useful, but still fall well short of top OpenAI and Anthropic systems for coding and reasoning.

The thread also pushed back on hype around cheaper open models, warning that benchmark claims can overstate real-world capability.