OpenAI cuts transcription costs by 25% — but trails Mistral and ElevenLabs

openai launched gpt transcribe and gpt live transcribe, processing audio 34x faster than real time and achieving a 3.31% word error rate on the aa-wer benchmark. pricing fell 25% to $0.0045 per minute, though mistral undercuts this at $0.003. despite improvements, openai ranks behind elevenlabs scribe v2 (2.3%) and google gemini 3 pro (2.9%) in accuracy. this release complements the realtime api as competition shifts toward faster, cheaper, and more precise speech models.