OpenAI has released a new suite of speech models, including GPT-Realtime-2, which features superior reasoning and an expanded 128,000 token context window. The update also introduces dedicated models for live translation and high-speed transcription to enable more complex voice-agent workflows.