OpenAI launches GPT-Live, doubling task completion rate and opening API access

OpenAI unveiled GPT-Live, a new voice-focused model, on September 11, 2026. The company said the model doubles task completion rates compared to the previous generation, achieves 92% comprehension accuracy, supports natural interruptions in conversation, and is now available on Genspark and via API.

Telnyx, a launch partner, announced the release in a social media post, highlighting GPT-Live's ability to capture nuance in voice AI, including tone, timing, emotion, and intent. Telnyx said its Voice API carries 16 kHz audio directly to the model without app-side resampling, preserving more of the original voice signal.

The company also published a guide on building outbound AI calling with GPT-Live and the Telnyx Voice API.