uhh self explanatory yk
1,000@EvanWang11j3 https://www.anthropic.com/claude-haiku-5-5
AA prbly resolves yes tho not outyet, released before oct 15th is YES, 5.5 ranked higher than gpt 6 astra prbly resolves NO, ranked above 5.1 on TB resolves NO
For the Sonnet 5.5 benchmark-regression answer, Anthropic’s official system-card summary looks like YES evidence: https://www.anthropic.com/transparency
The Sonnet 5.5 section explicitly summarizes its system card. Under “Honesty under pressure”, it reports worse MASK performance than Sonnet 5 and points to Section 6.3.2.2. MASK is expressly a benchmark of honesty under pressure: https://arxiv.org/abs/2503.03750
The answer says at least one benchmark in the system card, without limiting this to capability benchmarks. Could you resolve this answer, or point out a criterion I’m missing? I hold YES in this answer.