On Thursday, July 23, 2026, OpenAI and Anthropic each announced a voice update, but they are not the same feature. OpenAI brought GPT-Live voice to its desktop app to direct agents. Anthropic let Claude voice mode run on Opus, Sonnet and Haiku and use connected apps.
Key Takeaways
- OpenAI integrated its “GPT-Live” natural voice conversation model into the ChatGPT desktop app for Mac and Windows, enabling voice control of coding agent Codex and ChatGPT Work.
- Anthropic separately upgraded Claude’s voice mode, adding model-switching between Haiku, Sonnet, and Opus mid-conversation plus external tool integrations.
- Our reading: the two updates solve different problems, so choose by task. ChatGPT Voice is built to steer agents on a desktop; Claude voice mode gives spoken access to stronger models and connected apps on mobile, desktop and web.
What happened
OpenAI introduced GPT-Live, its new voice model family, for ChatGPT on July 8. On July 23 it added ChatGPT Voice to the macOS and Windows desktop app, where you can speak to start, check and steer tasks in Chat, Work and Codex.
OpenAI’s documentation lists Plus, Pro, Business, Edu and Enterprise plans.
The same day, Anthropic said Claude’s voice mode, previously limited to Haiku, now runs on Opus, Sonnet or Haiku, with a model picker you can use mid-conversation. Claude can also use connected tools such as Gmail, Google Calendar, Slack and Canva.
TechCrunch reported that Anthropic did not change the voice model itself.
Side by side: what each vendor shipped
| Item | ChatGPT Voice (OpenAI) | Claude voice mode (Anthropic) |
|---|---|---|
| Where it runs | macOS and Windows desktop app; iOS through Remote after pairing with a desktop host | Mobile, desktop and web, in beta; Anthropic says it works best on a phone |
| Plans | Plus, Pro, Business, Edu, Enterprise (desktop); Enterprise and Edu start with two weeks of early access | All plans; Free gets Haiku and one connected tool, paid plans get more models and tools |
| What voice can do | Start, check and steer tasks in Chat, Work and Codex; computer use; macOS screen context through Appshots | Talk with Opus, Sonnet or Haiku; switch models mid-conversation; use connected tools |
| Model choice | Runs on GPT-Live; in an existing Codex task it uses that task’s selected model | Starts from your last text-chat model; Claude Fable is not available in voice mode |
| Not available | Only one voice chat can be active across the desktop app at a time | Not in Claude Cowork or Claude Code, where only dictation works |
| Developer price | GPT-Live 1 in the API: $0.05 per minute for the voice layer since Sept 10 | Voice counts toward plan usage limits; we found no separate voice price |
Sources for the table: OpenAI’s ChatGPT Voice documentation, release post and API pricing page; Anthropic’s July 23 announcement and Voice help article; TechCrunch for the contrast between the two. Where a cell says we found nothing, that is an absence in the pages we read, not a vendor statement.
Two Lenses
Lens one: a market moving in one direction
Both companies improved their voice products within the same news cycle, which fits a shared push toward spoken interfaces. That is an inference from the timing, not something either company said.
Lens two: separate roadmaps
Both announcements came on Thursday, July 23, but they are not one release. GPT-Live reached ChatGPT on July 8, and TechCrunch described Anthropic’s update as coming weeks after it. We found no source showing that either company reacted to the other, so this post does not claim that.
What it costs
Neither consumer feature has a separate voice price in the documentation we read, and Claude says voice conversations count toward regular plan usage limits. OpenAI does price the voice layer for developers.
GPT-Live 1 reached the API on September 10 at $0.05 per minute, billed per second, with backend model and tool usage billed separately.
| Usage assumption | Voice layer at $0.05 per minute | Calculation |
|---|---|---|
| One 30-second session | $0.025 | 30 s ÷ 60 × $0.05 |
| One hour of conversation | $3.00 | 60 min × $0.05 |
| 20 minutes a day, 22 days | $22.00 | 440 min × $0.05 |
This is 2mind arithmetic on OpenAI’s list price, not a measured bill. It excludes the backend model and tool calls, which OpenAI bills separately and which can dominate the total. For why agent workloads use more tokens than chat, see our post on agent token costs.
Limits to check before you rely on either
- ChatGPT Voice: only one voice chat can run at a time in the desktop app, and screen context through Appshots works on macOS only and can be disabled by an organization.
- ChatGPT Voice inside an existing Codex task is described by OpenAI as rolling out, so it may not appear for every account or host yet.
- Claude voice mode: it does not work in Claude Cowork or Claude Code, does not detect your language automatically, and non-English support is in beta.
- Claude saves a text transcript of every voice conversation to chat history, and asks permission before using a connected tool.
- Both: we found no published reliability data for multi-step tasks directed by voice.
A decision checklist
- You want to speak to a coding or desktop agent, start tests, check progress and steer work: ChatGPT Voice in the desktop app is the one built for that, and it needs a paid plan.
- You want to think aloud with a stronger model and let it reach email, calendar or documents: Claude voice mode, which is on every plan, with Haiku and one connected tool on Free.
- You use a shared or sensitive device: check what each tool can reach (connected apps, screen context) and what is retained before turning voice on.
- You are building your own voice agent: price the $0.05 per minute voice layer plus the backend model separately, and test with your own tasks.
These rules follow from documented features, not from hands-on tests. For how to judge an agent’s reliability yourself, see Can You Trust AI Agents? Run the Same Task Ten Times.
Why it matters
Our reading: these are two different bets. OpenAI’s update treats voice as a controller for desktop agents, while Anthropic’s treats it as a conversation with a stronger model that can act in your apps. Whether either becomes a habit depends on usage evidence we could not find from either company.
FAQ
Q. Can I actually control coding tasks entirely by voice now with these updates?
A. Yes, in OpenAI’s desktop app. ChatGPT Voice can start, check and steer tasks in Chat, Work and Codex on macOS and Windows with a Plus, Pro, Business, Edu or Enterprise plan. We found no published reliability data for long multi-step tasks by voice, so start with low-risk work.
Q. Is Anthropic’s voice mode only available on paid Claude plans?
A. No. Voice mode is available to all users in beta, but free users are limited to the Haiku model and one connected app; the other models and more connected tools require a paid plan.
What would change our view
If OpenAI or Anthropic publishes error rates or reliability data for tasks directed by voice, the limits list above would change, and so would our advice to start with low-risk work.
If ChatGPT Voice reaches the free plan on desktop, or Claude voice mode gains support in Claude Cowork or Claude Code, the plan and platform rows of the comparison table would change.
If OpenAI changes the $0.05 per minute GPT-Live 1 price, the cost table above would be recalculated.
Sources
- TechCrunch — 2026-07-24. OpenAI integrated GPT-Live voice into ChatGPT desktop app enabling voice control of Codex
- TechCrunch — 2026-07-23. Anthropic upgraded Claude voice with model switching between Haiku, Sonnet, Opus mid-conversation
- SQ Magazine — 2026-07-24. Anthropic update adds external tool integrations to voice mode
- ChatGPT Voice (OpenAI documentation) — OpenAI, accessed 2026-10-02
- ChatGPT Voice is now in the desktop app — OpenAI Developer Community, 2026-07-23
- GPT-Live 1 model page and pricing — OpenAI, accessed 2026-10-02
- API changelog, Sept 10 entry for gpt-live-1 — OpenAI, 2026-09-10
- Think through hard problems in voice mode — Anthropic, 2026-07-23
- Use voice mode — Anthropic Help Center, accessed 2026-10-02
- OpenAI Introduces GPT-Live to Make ChatGPT Voice Feel Like a Real Conversation — MacRumors, 2026-07-08
Correction, October 2, 2026. An earlier version said Anthropic’s platform availability and plan limits were not specified. Anthropic’s voice update is in beta across platforms, and free users are limited to the Haiku model and one connected app.
Update, October 2, 2026. We revised this post. The title no longer says “this week.” Two graphics claiming a roughly 24-hour gap were removed because both announcements are dated July 23. The comparison table, limits, API price and checklist were added from official documentation.
Leave a Reply