AI Moves Beyond Chatbots: Robots, Voice Agents and Health Data

This week's most useful AI news happened outside the usual chatbot window.

Black Forest Labs put FLUX 3 into early access, with one model trained across image, video and audio. OpenAI added live voice controls to Work and Codex on desktop. Anthropic expanded what people can do with Claude Voice, and OpenAI opened ChatGPT Health to more users in the United States.

The products are very different, so it is worth separating what is available now from what is still being tested.

FLUX 3 combines image, video and audio

Black Forest Labs describes FLUX 3 as a multimodal foundation model trained across images, video and audio in the same architecture. The goal is to let the model learn how a scene looks, how objects move and how physical events sound, instead of treating each format as a separate task.

FLUX 3 Video can generate video with native audio from a text prompt, a starting image or a reference clip. Black Forest Labs says a single generation can run for up to 20 seconds. The model can also continue an existing video, create a transition between keyframes, handle multilingual dialogue and carry elements from a reference clip into a new scene.

Those features are in early access, and the company calls its evaluation results preliminary. That caveat matters. Black Forest Labs ran the comparisons itself, and the model is still changing during the early access period.

The robotics part is real, but limited

FLUX 3 also includes work on action prediction. Black Forest Labs gave mimic robotics early access to the video backbone, and the two companies developed FLUX-mimic for robot learning and dexterous manipulation. Their joint work is being tested on production tasks at Audi.

This is not a general robot controller that anyone can download today. Access to FLUX-mimic and FLUX 3 Action starts with selected research and commercial partners. FLUX 3 Image is due to enter early access later, while a planned FLUX 3 Dev release will offer open weights for the multimodal backbone.

Read the Black Forest Labs FLUX 3 announcement.

ChatGPT Voice can steer Work and Codex

OpenAI has added a live voice interface to Work and Codex in the ChatGPT desktop app. On eligible macOS and Windows accounts, a user can describe a task, ask for a progress update, interrupt the agent or change the instructions while work is underway.

The voice layer does not receive extra access by default. It can only use the files, desktop apps, tools and permissions available to the selected Work or Codex session. Some desktop actions may require microphone, screen recording or accessibility permission from the operating system.

There are a few practical limits. Only one Voice conversation can run at a time. OpenAI also says connected Voice time can be metered separately when flexible pricing applies, while tasks started through Voice use the same agent usage and credit pool as Work and Codex.

Work is available on web and mobile for eligible paid plans, but Voice for Work and Codex is a desktop feature. Codex itself remains a separate desktop view rather than a selectable experience on the web or mobile app.

Read OpenAI's Work and Codex guide.

Claude Voice is pulling in connected work data

The Superhuman newsletter reported that Claude Voice now supports Anthropic's Opus and Sonnet models. Anthropic's public voice guide does not yet document the model selector, so that detail should be treated as a reported rollout rather than a fully documented product change.

Anthropic does confirm that paid Claude users can bring Google Calendar and Gmail information into voice conversations. Google Docs access through voice is listed for Claude Enterprise. That means someone can ask about an upcoming meeting or recent email without leaving the spoken conversation.

Claude Voice is currently a mobile feature for iOS and Android, available in English, and voice messages count toward the user's normal usage limit. It is different from OpenAI's desktop agent control: Claude is using connected information inside a mobile voice conversation, while ChatGPT Voice can directly coordinate Work and Codex tasks on a computer.

Read Anthropic's Claude Voice guide.

ChatGPT Health reaches more U.S. users

OpenAI is gradually rolling out Health in ChatGPT to logged-in U.S. users aged 18 and older on Free, Go, Plus and Pro plans. It works on the web and iOS. Apple Health requires an iPhone, and users can also connect supported medical records from U.S. hospital systems, One Medical or Function Health.

Once a person gives permission, ChatGPT can use that information to compare a new lab result with earlier tests, summarize changes since an appointment or discuss patterns in sleep and activity. Health remains a dedicated area in the sidebar, but connected information can also be used in other conversations when the user allows it.

What OpenAI says about privacy

OpenAI says connected medical records, Apple Health information and conversations that use those records are not used to train its foundation models or target ads. By default, ChatGPT asks before using connected health information in a response. Users can switch to persistent permission or disconnect a source later.

Disconnecting a source starts deletion of its synced data from OpenAI's systems within 30 days. Information already copied into conversation history remains there until the user deletes those conversations. ChatGPT Health cannot update Apple Health or a provider's medical record, and it is not available in Codex.

OpenAI also states that Health is not intended to diagnose or treat a condition. The company tells users to verify important information and discuss medical decisions with a qualified professional.

Read OpenAI's Health announcement | Read the Health help guide.

What these releases have in common

Each update gives an AI system a new kind of context. FLUX 3 learns from several media formats at once. Voice lets people direct agents while a task is running. ChatGPT Health can read sensitive records that a user has chosen to connect.

That extra context can remove a lot of manual work. It also makes the limits easier to see. FLUX 3's robotics tools are still restricted, voice agents depend on the permissions they receive, and health answers still need human judgment. Those details will matter more to most users than another benchmark score.

Editorial note

The July 24 Superhuman newsletter surfaced these stories. Free Prompt Base checked the product details against the linked company documentation before publishing this expanded report. The cover is an original illustration, not a product screenshot.