- Claude voice mode now routes conversations through Opus or Sonnet rather than defaulting to the lighter Haiku model, unlocking longer reasoning and tool-heavy tasks via speech [1] - The voice interface includes a model picker and thinking-level control, with ElevenLabs handling text-to-speech output [1] - Anthropic has expanded Claude's workplace integrations to nine apps including Slack, Canva, Figma, Asana, and Box, all powered by the Model Context Protocol [2] - Voice mode supports 18 languages after multilingual input exited beta in June 2026, with push-to-talk and seamless text-voice switching [3] - The approach contrasts with OpenAI's speech-native bidirectional models, as Anthropic routes stronger reasoning models through conventional voice infrastructure
[1] Anthropic is upgrading Claude's voice mode to support its most powerful models — Opus and Sonnet — ending a period in which voice conversations were limited to the lighter, less capable Haiku model. The model picker in the voice interface, which had been visible for roughly three weeks but previously cosmetic, now routes conversations through whichever model the user selects [1].
The expansion unlocks a class of voice interactions that were impractical on Haiku, including tasks requiring extended reasoning chains and multi-step tool use handled entirely through speech. Anthropic continues to use ElevenLabs for text-to-speech output, keeping the same voice infrastructure while swapping in more powerful reasoning backends [1].
Alongside the model upgrade, Claude's ecosystem of workplace integrations has grown to include nine third-party apps — Slack, Canva, Figma, Asana, Box, Clay, Amplitude, Hex, and Monday.com — all connected through Anthropic's Model Context Protocol. The integrations are available to Pro, Max, Team, and Enterprise subscribers [2].
What Changed #
Until this update, Claude's voice mode ran exclusively on Haiku 4.5 regardless of which model users selected in the interface. The picker displayed Opus, Sonnet, and Haiku as options, but the selection was cosmetic — every session defaulted to Haiku. That limitation has now been removed, with voice sessions routing through the selected model [1].
A thinking-level control accompanies the model selector, letting users dial reasoning depth up or down during voice conversations. The voice interface also handles interruptions more gracefully, pausing when it detects user speech and resuming sensibly, while tolerating long silences without cutting the user off prematurely [1].
Anthropic's approach differs architecturally from OpenAI's voice strategy. Rather than building speech-native bidirectional models, Anthropic feeds its existing text-based reasoning models into a conventional voice pipeline with ElevenLabs handling the audio output layer [1].
Workplace Integrations #
The voice mode expansion arrives alongside a broader push into enterprise tool integration. Claude now connects to nine workplace platforms through the Model Context Protocol, the open standard Anthropic released in late 2024 and later donated to the Linux Foundation [2].
Through these integrations, users can draft and send Slack messages, create and edit Canva designs, generate Figma diagrams, build Asana project timelines, search Box documents, run Hex data queries, and manage Monday.com workflows — all without leaving the Claude interface. The integrations render interactive UI components inline rather than returning plain text [2].
A Salesforce integration is forthcoming. Google Workspace connectivity, including Gmail thread summarization, Google Calendar access, and Google Docs search, is available separately for Pro and Team plan subscribers [3].
Pricing and Availability #
Voice interactions on Opus or Sonnet count against users' standard message quotas, with no separate voice pricing tier. The feature is accessible across Claude's paid plans: Pro at $20 per month, Max at $100–$200 per month, Team at $25–$30 per user per month, and Enterprise at custom pricing [3].
Claude's voice mode now supports 18 languages after multilingual input exited beta in June 2026. Push-to-talk mode and smoother switching between text and voice input were added alongside the language expansion [3].
Claude Fable, Anthropic's creative-writing-focused model, is not available in voice mode at this time [1].
Competitive Context #
The upgrade narrows a capability gap with OpenAI's ChatGPT, which has offered voice conversations powered by its GPT-4o model since mid-2024. Google's Gemini also supports voice interaction with its most capable models across mobile and web interfaces.
Anthropic's decision to layer its strongest reasoning models onto existing voice infrastructure rather than training speech-native models represents a distinct technical bet. The tradeoff favors reasoning quality over conversational latency, positioning Claude voice as a productivity tool rather than a casual voice assistant.
The broader integration push — connecting Claude to Slack, Figma, Canva, and other workplace tools via MCP — reflects Anthropic's enterprise strategy of embedding AI into existing workflows rather than building standalone applications. Over 50 integrations are now available through Claude's connector directory [2].
Companies mentioned #
Further sources #
The stories that matter, in one email. Free — unsubscribe anytime.