Anthropic Upgrades Claude Voice Mode to Run on Sonnet and Opus
Claude's voice mode no longer runs solely on Haiku — paid users can now switch to Sonnet or Opus mid-conversation for complex spoken requests. Voice defaults to whichever model you last used in text.
Anthropic rewired Claude’s voice mode so it can run on Sonnet and Opus, not just Haiku. Until this update, every spoken conversation with Claude — regardless of how complex the request was — routed through Haiku, the smallest model in the lineup. That kept latency low but capped what voice mode could actually do: multi-step reasoning, nuanced coding questions, or anything requiring the depth Sonnet and Opus are built for got noticeably worse answers than the same question typed into chat.
Paid subscribers can now switch between Haiku, Sonnet, and Opus mid-conversation in voice mode, the same way they already do in text. Voice mode also picks up a smarter default: instead of always starting on Haiku, it now opens with whichever model you last used in a text chat, so a session that started on Opus for a hard problem stays on Opus when you switch to talking.
The change reflects where Anthropic has been pushing Claude as a product. By this point in 2026, “Claude” isn’t a single chatbot — it’s a family spanning chat, the API, Artifacts, web search, file handling, coding assistance, and computer use, with different models slotted in depending on the job. Voice was the one surface still locked to the cheapest, fastest model regardless of task complexity. Tying it to the same model tier a user is already paying for closes that gap.
It’s a small architectural change with an obvious practical effect: ask Claude a genuinely hard question out loud — debug a stack trace, reason through a contract clause, plan a multi-step task — and you now get Sonnet- or Opus-level reasoning instead of a fast-but-shallow Haiku answer. For anyone using voice mode as a hands-free coding assistant or research partner rather than a quick Q&A tool, that’s the difference between voice mode being a novelty and being usable for real work.
Separately, Anthropic is winding down Fast Mode for Claude Opus 4.7, with removal effective July 24, 2026 — the same day this voice update rolled out — as the company consolidates around its current model lineup.
No word yet on whether the model-switching behavior extends to free-tier users, who are more likely to stay capped on Haiku for cost reasons. For developers building voice interfaces on top of the Claude API, the underlying lesson is the same one Anthropic just applied internally: routing every request to the cheapest model by default trades away quality on the requests that need it most.