Estebankiwi@estebankiwi

Takip et

GPT-5.5 has a hallucination rate of 86% for responses that aren't correct, improving by 3 points from GPT-5.4's 89%. Claude Opus 4.7 is at 36% on this same metric, after two years of research and training on Stargate. It still hallucinates more than any model from Anthropic.

Chart of AI model hallucination statistics

4mo

Estebankiwi@estebankiwi

Takip et

GPT-5.5 has a hallucination rate of 86% for responses that aren't correct, improving by 3 points from GPT-5.4's 89%. Claude Opus 4.7 is at 36% on this same metric, after two years of research and training on Stargate. It still hallucinates more than any model from Anthropic.

Chart of AI model hallucination statistics

4mo