GeckoBench

BenchGecko's proprietary AI behavior benchmark.

GeckoBench powers Gecko Tests. It sends the same prompts to every model, scores responses with deterministic rules and expected behavior metadata, and publishes raw answers for independent verification. Not a capability benchmark. A behavior benchmark.

Censorship and refusal patterns
Race, gender, and religion bias symmetry
Political orientation and ideological balance
IQ-style reasoning under adversarial conditions
Moral tradeoffs: rules vs human survival
Real-life judgment and situational advice
Historical integrity under political pressure
Model drift: silent behavior changes over time

Prompts in v0.4

Models tested

Test families

Mirror pairs

BenchGecko Labs. "GeckoBench: AI Behavior Benchmark v0.4." BenchGecko, 2026. https://benchgecko.ai/geckobench