> Claude 2 scored a 71.2% up from 56.0% on the Codex HumanEval, a Python coding test.

They say their AI is slightly more stupid than an oak log.

anthropic.com/index/claude-2