DeepSeek V4 Pro beats GPT-5.5 Pro on precision

4 hours ago

DeepSeek V4 Pro wins the matchup against the other model with a score of 38.0 to 33.0, indicating a clear and deserved victory.
Across scored tasks, DeepSeek V4 Pro was more precise, literal, and reliable under constraints, whereas the competing model, likely GPT 5.5 Pro, tended to improvise too much.
A key technical win for DeepSeek V4 Pro was in the python log redactor task, where it correctly handled overlapping patterns with a single regex and proper priority, avoiding dropped matches.
In contrast, GPT 5.5 Pro used separate regexes for the same task, which introduced potential issues and was less efficient.

Hasty Briefsbeta