Even the best LLMs are bad at programming

• June 17, 2025 at 14:27 IST
• coding
• llm
• ai

As per this X post about LiveCodeBench, almost all LLMs are bad at competitive programming. At the time of writing this, the o3-high (2025-04-16) model was the best at medium difficulty programming (which I don't believe, but haven't really used o3 a lot, so I'd like to believe it).

But I do agree that none of the models are good at hard difficulty programming tasks. I have used them extensively and, while they are a great assistant, they are not good at complex tasks.

• ← Previous
  DHH about seasons of saying NO and YES
• Next →
  MCP containers