OpenAI released two new transcription models on July 29: GPT-Live-Transcribe for real-time low-latency use at $0.017 per minute and GPT-Transcribe for asynchronous batch processing at $0.0045 per minute. Both models understand context including accents, terminology, numbers, and background noise. On the Context Aware ASR benchmark, GPT-Transcribe improved semantic accuracy from 41.6% to 45.2% with context. Compared to Whisper, GPT-Transcribe reduced error rates to 19.27% vs 40.37% on Common Voice and 8.98% vs 15.21% on real-world recordings.