Anthropic updated Claude's voice mode on July 23, 2026, so that spoken queries can be handled by Sonnet and Opus rather than only Haiku, the smallest model, which previously handled all voice traffic to keep latency low. Paid users get voice mode defaulting to whichever model they last used in text chat, and can switch between Haiku, Sonnet and Opus mid-conversation via the model picker; Anthropic says voice mode uses the fastest version of the selected model. Voice mode can now also pull context from connected apps such as Gmail and Slack with user permission. The architecture remains turn-based rather than full duplex like OpenAI's GPT-Live, and language switching mid-conversation must be declared manually, though more languages including Indonesian were added. The beta is rolling out to all users on desktop, mobile and web; free accounts are limited to one connection and Haiku only.