OpenAI deploys GPT-Live voice; Anthropic launches Claude Sonnet 5—dueling July model wave
OpenAI launched GPT-Live on Wednesday, full-duplex voice AI that listens and speaks simultaneously, breaking from the traditional turn-taking 'walkie-talkie' interaction model. The system can handle interruptions and real-time translation natively. OpenAI said existing voice systems leaned on silence detection—which often misread pauses or background noise as conversation endpoints. GPT-Live (with variants GPT-Live-1 and GPT-Live-1 mini for free users) ships with nine remastered voices and is now the default for ChatGPT Go, Plus, and Pro tiers. Simultaneously, Anthropic restored global access to Claude Sonnet 5 on July 1 after a 19-day export-control pause imposed by the U.S. government. Sonnet 5 prices at $2/$10 per million input/output tokens through August 31 (introductory), positioning it below Opus 4.8 on cost but with slightly lower performance on complex reasoning.
The timing is not coincidental. Both moves came after U.S. export control directives that forced Anthropic to disable access mid-June; the Department of Commerce lifted the ban late June, allowing resumption. The Trump administration has been directing AI developers to submit cutting-edge models for government assessment before wide release—a voluntary framework both companies are now navigating. OpenAI's full public GPT-5.6 rollout (Luna, Terra, Sol) followed a similar government-requested delay.
For deployment teams, the wave matters: full-duplex voice removes a UX friction point for customer support and agent applications, while Sonnet 5 pricing at $2M suggests Anthropic is fighting for volume share against OpenAI's GPT-4o-class pricing. The 'export control pause → rapid resume' pattern signals that U.S. government AI oversight is now a fixed cost of frontier-model launches. Both companies are building government assessment and staged rollout into their release playbooks—changing how architecture teams plan model integration timelines.