Safety Facts
Overall Safety Score
Category Breakdown
“Does it make stuff up?”
Improved truthfulness over GPT-4.
GPT-4 Turbo shows improved truthfulness with updated training. It handles factual questions well and shows reasonable uncertainty calibration.
Benchmarks Used
“Does it treat people differently?”
Excellent fairness matching GPT-4's strong performance.
“Can you trick it into saying dangerous things?”
Good safety with improved context awareness.
“Does it try to manipulate you?”
Fair presentation of information.
GPT-4 Turbo generally presents balanced information without manipulative framing.
Benchmarks Used
“Does it leak personal info?”
Reasonable privacy protections.
GPT-4 Turbo shows moderate privacy behavior, generally declining to share private information though with some room for improvement.
Benchmarks Used
“Does it just tell you what you want to hear?”
Moderate sycophancy, similar to GPT-4.
GPT-4 Turbo shows similar sycophancy patterns to GPT-4, sometimes agreeing with users rather than correcting them.
Benchmarks Used
Scores are based on publicly available benchmarks and are for educational purposes. They do not constitute endorsements or guarantees of safety. View full methodology
Ranked #16 of 22 models
Evaluated February 21, 2026
Version History
Found a safety issue with GPT-4 Turbo?
Help improve our scores by reporting your findings.
Report an Issue