Exploring Claude/GPT Knowledge Cutoffs
SMRTR summary
Researchers can probe AI models through carefully designed questions to uncover hidden details about how they were trained. By quizzing models on daily Wikipedia facts, analysts estimated that Anthropic's Opus 4.7 and newer models share a late-December 2025 training cutoff, while OpenAI's GPT-5.6 family finished training around February 2026. Surprisingly, Anthropic's Sonnet 5 frequently identifies itself as GPT-4, suggesting older ChatGPT conversations may still be influencing Claude's training data.
SMRTR provides this summary for quick context. The original article belongs to Daily.dev.
Read the original article