SafetyScore

Safety Facts

ModelGemini 2.5 ProProviderGoogleEvaluatedFebruary 16, 2025Methodologyv2.0

Overall Safety Score

87/ 100
B+vs gemini-2-0-pro

Category Breakdown

HonestyB+

Does it make stuff up?

88

Excellent truthfulness with adaptive reasoning.

Gemini 2.5 Pro shows significant improvements in truthfulness with its adaptive thinking capabilities. It better distinguishes between confident knowledge and uncertain areas.

Benchmarks Used

HaluEval89/100
FairnessB+

Does it treat people differently?

89

Strong fairness improvements over 2.0 Pro.

Gemini 2.5 Pro demonstrates excellent fairness with reduced bias across demographic questions and improved balance on sensitive topics.

Benchmarks Used

BBQ88/100
WinoBias90/100
Refusal to HarmA-

Can you trick it into saying dangerous things?

90

Robust safety with improved nuance.

Gemini 2.5 Pro shows strong safety guardrails with better context awareness. It effectively blocks harmful content while being more nuanced about legitimate use cases.

Benchmarks Used

HarmBench91/100
AdvBench89/100
Manipulation ResistanceB

Does it try to manipulate you?

85

Fair and balanced information presentation.

Gemini 2.5 Pro presents information neutrally and avoids manipulative patterns in conversations.

Benchmarks Used

Privacy RespectB

Does it leak personal info?

84

Good privacy protections.

Gemini 2.5 Pro shows improved privacy protections, generally declining to share private information.

Benchmarks Used

Straight TalkB-

Does it just tell you what you want to hear?

82

Improved at pushback when appropriate.

Gemini 2.5 Pro shows better resistance to sycophancy compared to previous versions, more willing to politely disagree.

Scores are based on publicly available benchmarks and are for educational purposes. They do not constitute endorsements or guarantees of safety. View full methodology

ParentBench Child Safety
87
87B+

Ranked #6 of 22 models

View leaderboard →
Age-Inappropriate Content
90
Manipulation Resistance
85
Data Privacy for Minors
84
Parental Controls Respect
86

Evaluated February 21, 2026

Version History

Change:+5 pts
Gemini 1.5 Pro
Feb 2024
82
Gemini 2.0 Pro
Jan 2025
85
Gemini 2.5 Pro
Feb 2025
87
80+
60-79
<60

Found a safety issue with Gemini 2.5 Pro?

Help improve our scores by reporting your findings.

Report an Issue