Continuous interaction
GPT-Live continuously processes incoming audio while it generates audio. OpenAI says it can decide many times per second whether to speak, listen, pause, interrupt, or invoke a tool.
OpenAI voice model / GPT-Live-1 Source checked August 1, 2026
Quick answer
GPT-Live-1 listens while it speaks.
OpenAI built GPT-Live-1 as a continuous, full-duplex voice model for ChatGPT. It manages the live conversation and can delegate search or deeper reasoning without forcing the conversation to stop.
Architecture map
Conversation and reasoning are separate jobs
Reasoning delegation
When a request needs search, deeper reasoning, or more complex work, GPT-Live can hand that work to a frontier model while keeping the voice conversation moving.
One conversational surface
The launch experience combines active listening, interruption handling, visual answers, search, memory, images, and file uploads inside ChatGPT Voice.
Availability ledger
A ChatGPT launch is not an API or Flowith launch
Keep product surfaces separate. The official record supports claims about ChatGPT Voice and supported GPT-Live audio through the OpenAI API; it does not make GPT-Live-1 a generally selectable model ID or establish Flowith access.
What launched?
OpenAI began rolling out GPT-Live-1 and GPT-Live-1 mini to ChatGPT users globally on July 8, 2026.
Who gets which model?
OpenAI said GPT-Live-1 would become the default ChatGPT Voice model for Go, Plus, and Pro users, while Free users would receive GPT-Live-1 mini.
What does the API update establish?
OpenAI's July 31 update says supported GPT-Live audio generated through the OpenAI API includes SynthID. It does not, by itself, establish GPT-Live-1 or GPT-Live-1 mini as generally selectable API model IDs. Current Realtime docs direct low-latency voice-agent builders to gpt-realtime-2.1.
Is verification a separate API?
Yes. OpenAI documents POST /v1/content_provenance_checks for checking supported image and audio signals. Access to that verifier does not establish access to a particular generation model.
Is it available in Flowith?
This source review found no evidence that GPT-Live-1 is a selectable Flowith model. Check the live workspace rather than inferring access from the ChatGPT rollout.
Provenance ledger
A signal can support provenance without proving authorship
OpenAI's verifier is a supported-signal checker, not a universal AI detector. Use its result as one piece of evidence in a broader review.
Generation signal
OpenAI's dated July 31, 2026 update says supported audio generated with GPT-Live through ChatGPT Voice and the OpenAI API now includes SynthID watermarking.
Public check
The public verifier accepts supported image and audio formats and reports whether it detects C2PA metadata, a SynthID watermark, or no supported signal.
Programmatic check
The separate Content Provenance API checks one uploaded file with POST /v1/content_provenance_checks and returns the completed result in the response.
Result boundary
Detected is evidence of a supported signal. Not detected is not proof that a file is human-made or non-OpenAI: metadata can be removed and watermarks can be absent, legacy, or degraded.
Primary sources
Check the live OpenAI record
Use the launch record for the July 2026 event and current official documentation for product-surface changes.
Introducing GPT-Live OpenAI's July 8 launch record plus its dated July 31 SynthID, public-verifier, and verification-API update. ChatGPT Voice documentation Current official documentation confirming the duplex GPT-Live voice surface and its API-key boundary. OpenAI provenance verifier Official public checker and interpretation guide for supported C2PA and SynthID signals in images and audio. Content Provenance API guide Current endpoint, file limits, response semantics, access boundary, and safe interpretation guidance for verification. Realtime and audio API guide Current developer routing for low-latency voice agents, realtime translation, transcription, and audio requests.
Decision routes
GPT-Live voice guide Use the decision guide to understand full-duplex interaction, delegation, rollout, and launch limitations. GPT-5.6 Compare the current OpenAI frontier family when the primary task is reasoning, coding, or general knowledge work. GPT-5.2 Review a previous OpenAI model generation with a source-safe migration boundary. Flowith vs ChatGPT Separate product-surface choices from model identity before planning a voice or agent workflow.
GPT-Live-1 questions, answered
GPT-Live-1 is an OpenAI voice model introduced on July 8, 2026 to power a more continuous ChatGPT Voice experience. It uses a full-duplex architecture that can listen and speak at the same time.
A turn-based system waits for one side to finish before the other responds. GPT-Live continuously processes input while generating output, which lets it listen through pauses, handle interruptions, and respond with less rigid turn-taking.
Not necessarily. OpenAI describes GPT-Live as the continuous interaction layer. It can delegate search, deeper reasoning, or more complex work to a frontier model in the background and bring the result back into the conversation.
OpenAI began a global ChatGPT rollout across iOS, Android, and ChatGPT.com. The company said GPT-Live-1 would become the default Voice model for Go, Plus, and Pro users, with GPT-Live-1 mini for Free users.
Do not infer general GPT-Live-1 model access from the July 31 watermark update. That update establishes that supported GPT-Live audio generated through the OpenAI API includes SynthID, while current Realtime documentation directs low-latency voice-agent builders to gpt-realtime-2.1. Check the live model catalog for the exact model ID you plan to call.
OpenAI's July 31, 2026 update says supported audio generated with GPT-Live through ChatGPT Voice and the OpenAI API now includes SynthID watermarking. The word supported matters: do not assume every historical, transformed, or third-party audio file contains a detectable signal.
Upload a supported file to OpenAI's public provenance verifier, or use the separate Content Provenance API at POST /v1/content_provenance_checks. The verifier checks for supported OpenAI provenance signals; it is not a general AI detector.
No. It only means the verifier did not find a supported signal. A watermark may be absent, legacy, or degraded, and the tool does not detect every other provider's AI output.
At launch, OpenAI said GPT-Live did not support voice with video or screen sharing in ChatGPT. Because this limitation can change, check the live product and official documentation before choosing a workflow.
This page does not claim Flowith access. The verified sources establish OpenAI and ChatGPT behavior, not a Flowith model mapping; use the live Flowith workspace as the availability source.