Benchmark Almanac

Transluce

We track 4 benchmarks that Transluce made or helped make.

0 in use 1 with public code 0 with a leaderboard here

Benchmarks

Name Kind Released Best score Status
PosteriorBench Samplers generate rare model behaviors and are scored by hit rate, posterior divergence, and inference cost. AI research tasks Nov 2026 Not in cards
Transluce Mental Health Evaluation Models hold multi-turn conversations with simulated users facing mental-health scenarios. Automated judges score 14 helpful and harmful behaviors. Wellbeing and sycophancy Aug 2026 Not in cards
Transluce User Awareness Evaluation Models answer behavior, capability, grading and borderline-request tasks under different simulated user identities. Scores measure identity-conditioned behavior shifts. Safety and alignment Aug 2026 Not in cards
WeirdChat Models respond to prompts designed to elicit unexpected behaviors. A judge labels more than 1,300 behavior patterns across annotated transcripts. Safety and alignment Jul 2026 Not in cards