Anthropic Upgrades Claude Voice Mode with Advanced AI Models and App Integration
Artificial intelligence research company Anthropic has released a beta update to the voice mode feature in its Claude assistant, granting spoken interactions access to the company’s most sophisticated frontier models, Sonnet and Opus, as well as live data from connected third-party applications.
The enhancement significantly expands the capability of Claude’s spoken interface, which previously relied exclusively on Haiku, Anthropic‘s smallest and fastest AI model designed primarily to minimize response latency. While Haiku handled routine spoken queries efficiently, complex reasoning tasks often exceeded its underlying processing capacity. Under the new system, paid subscribers can utilize Sonnet or Opus depending on their needs, with voice conversations defaulting to whichever model they last selected in their text chat session. Users can also toggle between Haiku, Sonnet, and Opus mid-conversation using an updated model selection tool.
Anthropic’s model portfolio is structured around distinct performance tiers. Haiku is optimized for rapid, cost-effective responses; Sonnet serves as an enterprise workhorse balancing processing power and speed; and Opus represents the firm’s highest tier of intelligence, engineered for multi-step logic and intricate analytical work. By enabling Sonnet and Opus in voice mode, Anthropic aims to bridge the functional gap between typing complex prompts and executing them verbally.
Alongside model flexibility, the updated voice mode introduces integration with external productivity platforms. Provided users grant explicit permission, Claude can now pull contextual information from connected services such as Gmail and Slack during verbal sessions. This shift aligns with a broader technology industry drive toward agentic AI systems—assistants capable of reading workspace data and executing cross-platform workflows rather than serving solely as conversational search tools.
Despite these additions to intelligence and tool access, Anthropic’s voice architecture remains distinct from some of its primary competitors. The system operates on a turn-based model, where Claude listens to a full user statement, pauses to compute a response, and then speaks. This differs from full-duplex conversational frameworks, such as OpenAI‘s GPT-Live or Advanced Voice Mode, which process incoming audio continuously and allow natural mid-sentence interruptions.
Language handling in Claude’s voice feature also retains specific procedural requirements. The assistant cannot automatically detect when a speaker switches languages during a conversation. To transition between languages, users must either instruct the system vocally before speaking in the new language or select it manually within the voice settings menu. Anthropic has, however, expanded overall language availability in this update, adding support for several new languages, including Indonesian.
The rollout applies across Claude’s web client as well as its desktop and mobile apps. While paying customers gain access to model switching and workspace integrations, users on free accounts remain restricted to a single connection routed through Haiku, though they can access all newly supported languages. Anthropic stated that the current release prioritizes intelligence and integration capabilities, with further developments in voice technology planned for later this year.
