GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 -2.3%
GPT-6A 820 —
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 586 -0.5%
INKL 531 —
CL-OP46 497 —
CL-OP48 490 -0.2%
GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 -2.3%
GPT-6A 820 —
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 586 -0.5%
INKL 531 —
CL-OP46 497 —
CL-OP48 490 -0.2%
← Back to feed

Anthropic Expands Claude Voice to All Three Model Families With Cross-App Actions

Anthropic has upgraded Claude’s voice mode to support all three model families. The previous implementation ran only Claude Haiku — fast, but unsuitable for sustained or complex reasoning in spoken form. The update makes Opus, Sonnet, and Haiku available for voice conversations, switchable mid-session without losing conversational context between speaking and typing.

The second change is more functionally significant: voice mode now executes actions across connected applications. Gmail, Google Calendar, Slack, Canva, and Notion are in the initial set. After requesting permission, Claude can reschedule meetings, draft email replies, summarize Slack threads, and create Notion documents through voice alone. The permission gate is required per action.

What Changed in the Model Layer

Anthropic changed the reasoning layer, not the underlying speech stack or interruption behavior. Latency characteristics on Haiku remain the same. Opus and Sonnet carry higher inference costs, which is why free accounts remain limited to Haiku and a single connected tool. Paid tiers on mobile, desktop, and web can access all three families and the full tool set.

The connected-app model requires the assistant to identify intent from speech, match it to a tool, surface a permission request, and execute only after confirmation. That flow is more complex than a text-based tool call because the intent boundary is less precise in spoken language and the interruption window is shorter.

Why Haiku-Only Was a Ceiling

The original voice mode ran Haiku because voice demands low latency above almost everything else. Haiku has the fastest time-to-first-token in the Claude family and the smallest footprint per call, which keeps spoken responses from feeling delayed. The tradeoff was capability: Haiku handles conversational exchanges and simple lookup well, but multi-step reasoning tasks — planning a trip, restructuring a document, working through a legal summary — are better suited to Sonnet or Opus.

Allowing model selection mid-conversation addresses this without requiring users to restart. A session that begins with casual calendar queries can escalate to Opus for a complex analytical request and return to Haiku without the conversation resetting.

Language and Access

Voice mode now supports 10 languages in the beta. The rollout reaches free and paid users across mobile, desktop, and web. Free accounts are capped at Haiku and one connected tool; paid tiers get full model selection and the complete connected-app set.

The Slack, Canva, and Notion connections extend Claude’s reach into collaboration and creative tools beyond the productivity layer that Gmail and Calendar represent. The pattern matches what enterprise customers have been building manually through Claude’s API: a single spoken interface that can move work across the applications where it actually lives.

Key Numbers

  • Model families in voice: Opus, Sonnet, Haiku (previously Haiku only)
  • Connected apps: Gmail, Google Calendar, Slack, Canva, Notion
  • Languages: 10
  • Free tier: Haiku + 1 connected tool
  • Paid tier: full model selection + all connected apps
  • Platforms: mobile, desktop, web