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