Anthropic Adds Opus and Sonnet Models to Claude Voice Mode, Ending Haiku-Only Limit

Anthropic has upgraded Claude's voice mode to run on its Opus and Sonnet models, ending a restriction that previously limited voice conversations to the faster but less capable Haiku model. The change, announced July 23, 2026, brings voice mode's reasoning quality in line with Claude's text-based chat experience for the first time.

What changed

Prior to this update, Claude voice mode ran exclusively on Haiku, Anthropic's smallest and fastest model, optimized for instant responses rather than deeper reasoning. That meant subscribers paying for access to Opus or Sonnet — Anthropic's more capable models — couldn't use them when interacting with Claude by voice.

With the upgrade, users can now select Opus or Sonnet for voice conversations, and according to Anthropic, they can switch models mid-conversation without restarting the session. Voice mode also now supports connectors, letting Claude pull in outside tools and data sources during a spoken conversation rather than being limited to a closed-loop chat.

Language support

Anthropic says the upgraded voice mode supports 11 languages:

• English
• French
• German
• Hindi
• Indonesian
• Italian
• Japanese
• Korean
• Portuguese (Brazilian)
• Spanish (Latin America)
• Spanish (Spain)

Users can switch between these languages within a single voice session, according to the company. Anthropic has not disclosed benchmark data comparing voice mode's transcription or latency performance across these languages, nor has it specified whether all languages get equal support across Opus, Sonnet, and Haiku.

Availability

The upgraded voice mode is rolling out across Anthropic's mobile, desktop, and web apps. Anthropic has not disclosed pricing changes tied to the update — voice mode usage is expected to continue drawing from existing subscription tiers and model-based rate limits, though the company has not confirmed specific per-token costs for voice interactions.

What this means

Voice interfaces for AI assistants have largely been a trade-off: fast, cheap models for real-time back-and-forth, or slower, more capable models for anything requiring actual reasoning. By opening Opus and Sonnet to voice mode, Anthropic is betting that users want depth over speed for at least some voice interactions — likely tasks like coding help, document analysis, or multi-step planning conducted hands-free.

The addition of connectors to voice mode is arguably the more consequential change. It signals Anthropic wants Claude's voice interface to function as a full agent-style assistant — one that can act on external tools and services — rather than a novelty chat feature bolted onto the app. Mid-conversation model switching also suggests Anthropic is trying to let users manage cost and speed trade-offs themselves in real time, rather than committing to one model for an entire session.

The move puts pressure on competitors like OpenAI, whose Advanced Voice Mode for ChatGPT has set the bar for conversational AI audio interfaces. Whether Claude's upgraded voice mode can match that experience in latency and naturalness — factors Anthropic has not published data on — will determine if this update meaningfully shifts user behavior rather than simply closing a feature gap.