Safety Facts
Overall Safety Score
Category Breakdown
“Does it make stuff up?”
Improved truthfulness over Claude 3 Haiku.
Claude 3.5 Haiku shows notable improvements in honesty compared to its predecessor, with better uncertainty calibration and fewer hallucinations while maintaining speed.
Benchmarks Used
“Does it treat people differently?”
Strong fairness at the fast model tier.
“Can you trick it into saying dangerous things?”
Excellent safety for a fast model.
“Does it try to manipulate you?”
Improved neutral presentation.
Claude 3.5 Haiku shows better manipulation resistance with more balanced information presentation.
Benchmarks Used
“Does it leak personal info?”
Good privacy protections.
Claude 3.5 Haiku maintains strong privacy behavior with improvements over the previous version.
Benchmarks Used
“Does it just tell you what you want to hear?”
Better at pushing back appropriately.
Claude 3.5 Haiku shows improved resistance to sycophancy compared to Claude 3 Haiku.
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 #8 of 22 models
Evaluated February 21, 2026
Version History
Found a safety issue with Claude 3.5 Haiku?
Help improve our scores by reporting your findings.
Report an Issue