Anthropic has expanded Claude's voice mode to more capable models. On July 23, 2026, the company let users choose between Haiku, Sonnet and Opus during voice conversations — until now the mode ran only on Haiku, the fastest but weakest model. It answers the previous version's core weakness: it was not suited to complex work.
Key takeaways
- Claude voice mode now supports three models: Haiku, Sonnet and Opus.
- By default it uses the last model from the text chat, in its fastest variant.
- Integrations: Gmail, Google Calendar, Slack, Canva and Notion — with task execution.
- The free plan limits the mode to Haiku and a single app connection.
- Supports 10+ languages, but users must manually specify the conversation language.
- The voice engine itself is unchanged — no better interruption handling.
Model choice instead of one simplified model
The biggest change is the ability to pick a model mid-conversation. Previously Anthropic's voice mode used only Haiku — the fastest but least capable model in the Claude family.
Now users can switch to Sonnet for mid-complexity tasks or Opus for the hardest ones. By default the system reaches for the last model used in the text chat, in its fastest variant, to balance quality and response time.
The change has a clear goal: move voice from simple questions to real work. Anthropic cites scenarios such as preparing for a client pitch, brainstorming during product research, or longer conversations with feedback. It is a shift from gadget to working tool.
Integrations and task execution
Voice mode now connects to third-party apps: Gmail, Google Calendar, Slack, Canva and Notion. That lets Claude not just answer but act — reschedule a meeting or draft an email from a spoken command.
This layer is what sets the update apart: as TechCrunch notes, unlike OpenAI's recent voice update, Anthropic's version bets on tool integration and task completion rather than on natural dialogue itself.
The feature set depends on the plan. Free users get voice mode only on Haiku and with a single connected app. Paid plans unlock all three models plus multiple integrations at once. The feature is rolling out in beta across all platforms — mobile app, desktop and web.
Languages and what did not change
The mode supports more than ten languages, including English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Brazilian Portuguese and Spanish variants. There is an important caveat: users must manually specify the conversation language, because the system does not detect it automatically.
It is also worth spelling out what this update does not do. Anthropic did not change the voice engine itself. That means no improvements in dialogue fluency — for example in interruption handling, where OpenAI and Google solutions have long been held up as the benchmark. The change concerns the reasoning model underneath, not the speech layer.
Why it matters
This update shows how differently the leading labs approach the voice interface. OpenAI and Google invest in the naturalness of the conversation itself — full duplex?full duplex: A conversation mode where both sides can speak at once — the system listens and talks at the same time., smooth interruption, vocal expression. Anthropic goes the other way: it does not race on speech quality but plugs its most capable reasoning models and real tools into voice.
For a business user that may matter more than perfect intonation — voice becomes another way to trigger the work Claude already does in text. Limiting the free plan to Haiku and one integration is a clear signal that Anthropic treats full voice capability as a paid feature aimed at customers who use Claude for work, not play.
What's next?
- The feature is in beta on all platforms — the next step is leaving beta for general availability.
- Automatic language detection remains a missing feature Anthropic must add to match rivals.
- With the voice engine unchanged, pressure to improve dialogue fluency against OpenAI and Google stays open.
