davispalmie·2 年前·議論Command execution and formatting in the evals is nice instead of just tunneling on transcription accuracy. Are the new benchmarks open sourced?