Claude's voice mode used to be a quick-questions tool. You'd ask something on the drive in, get a fast answer, and move on to the next thing. From 23 July 2026, Anthropic changed what's underneath it. Voice mode now runs on Claude Opus and Claude Sonnet as well as Haiku, and it can act on tools you've already connected, such as Gmail and Slack. It also speaks 11 languages, including Hindi, Indonesian, Japanese, Korean, and both Latin American and European Spanish.
For a Sydney business owner talking through a problem between meetings, or working through a decision on the drive in, that's a real shift, not a cosmetic one. Before, voice mode gave you a fast answer from a small model built for speed rather than depth. Ask it something that actually required thinking through, and it hit a wall quickly. Now you can pick the model mid-conversation, and Claude keeps using the fastest version of whatever you've chosen, so the conversation still feels quick even when the thinking behind it isn't shallow.
What changed, and why the model matters
The old version of voice mode was fine for simple questions. Ask it to check tomorrow's weather or confirm a meeting time, and it did the job without fuss. Ask it to help you think through a genuinely hard decision, and it hit a ceiling fast, because the model behind it wasn't built to reason at depth in the first place. Swapping in Opus and Sonnet changes that ceiling. You're no longer trading depth for speed every time you open your mouth instead of your laptop. You get to choose which one you need, in the moment, without switching apps or typing a word, and Claude handles the rest.
The language support matters more than it might look on paper too. A business dealing with suppliers, contractors, or clients who think more naturally in Hindi, Japanese, Korean, Indonesian, or Spanish now has one tool that works the same way regardless of which language the conversation starts in. That's a small detail on a feature list, but it's the kind of thing that quietly removes friction from a working day.
There's also a practical side to this that's easy to miss: you don't need to know in advance which model you'll want. Start a conversation assuming Haiku speed, and if the topic turns out to need more depth, ask for Opus or Sonnet mid-stream instead of restarting from scratch. That's a small workflow change, but it removes the guessing game of picking the right tool before you've even framed the question.
What it's actually good for
Anthropic is pitching voice mode as a sounding board, somewhere to practise a pitch, decide between two supplier quotes, or talk out loud through a decision you keep circling back to. That's useful on its own. But in our experience running Claude Cowork setups for Australian businesses, the bigger shift is what happens after the talking stops. A few examples of what that looks like in practice:
Ask Claude to push a meeting back 30 minutes in your calendar, hands-free, while you're still in the car.
Turn a client-pitch conversation into a one-page brief without opening a laptop.
Get a summary of today's inbox and draft replies to the messages that actually need one.
Talk through a few competing ideas for a product or service change and have Claude sanity-check them against what's already out there.
None of this replaces judgment, and Claude still asks permission before it touches anything connected to your accounts. The change is that voice mode now does the thinking and the doing, not just the talking, which is a different kind of tool than the one most people tried a year ago.
Why this matters in dollar terms
A part-time executive assistant in Sydney costs somewhere around $35 to $45 an hour for exactly the admin work this replaces: calendar shuffling, drafting routine replies, chasing a conversation into an actual action item. Voice mode isn't a replacement for that person's judgment, and it's not trying to be. It's a replacement for the mechanical half of the job, the parts that eat an afternoon without needing much of a decision made along the way. The model doing that work is now capable enough to be worth trusting with it.
Run the numbers against your own week rather than ours. If voice mode picks up even a slice of the calendar admin, inbox triage, and note-to-brief work that would otherwise sit with an EA billing $35 to $45 an hour, the case for testing it properly starts to look less like a novelty and more like a line item worth reviewing at the next budget conversation. For a business owner or a small team without dedicated admin support, that's often the gap costing the most hours and getting the least attention, simply because nobody's had time to fix it.
The catch
Voice mode connecting to Gmail and Slack is genuinely useful, and it's also the point where a business needs to be deliberate. Which tools are connected, what Claude is allowed to do without asking first, and where a person needs to stay in the loop matters more as the model gets more capable, not less. If you handle client data, financial records, or anything that touches your obligations under the Australian Privacy Act, that's worth a proper look before you connect anything, not after the fact.
This isn't a reason to avoid it. It's a reason to set it up properly the first time, the same way you'd set access levels for a new staff member rather than handing over every password on day one.
The update is in beta now on mobile, desktop, and web. Free plan users get Haiku, one connected tool, and all 11 languages. Paid plans get the full model range and every tool you've connected. Either way, the conversation about what's safe to hand over takes five minutes and it's a lot cheaper than sorting out the mess of getting it wrong later.
If you want a plain-English read on what's safe to hand voice mode versus what should stay manual for your business, that's exactly the kind of thing we scope in an Automata AI Claude Cowork setup. Book a brainstorm session and we'll walk through where voice mode fits into your week, and where it doesn't, before you connect a single tool.
