Last week, OpenAI released its first hardware — a fancy little keyboard built to work with ChatGPT. Developed in collaboration with specialty keyboard designer Work Louder, the Micro is essentially a workplace innovation that many tech enthusiasts will love, but may leave everyone a little surprised.

OpenAI’s entry into the hardware market has not been without drama. A few weeks ago, Apple sued the artificial intelligence lab and accused it of commercial theft, in what is sure to be a long legal battle. Meanwhile, news of another shrewd home product in development at OpenAI also raised eyebrows, as the purported device – which is said to connect to ChatGPT – was reportedly developed by former Apple engineers.

Until the legal battle ends and OpenAI’s broader hardware ambitions materialize, the only thing the startup has to offer is the Micro, a tiny, funky keyboard clearly designed to delight the tech industry’s coding monkeys.

OpenAI sent a Micro test unit to TechCrunch. The first large thing you notice when you initially exploit the keyboard is that it’s a tiny, sturdy device — so much so that if the AI ​​accessory doesn’t work, it can easily double as a paperweight. Another thing you may notice (and I’m not only one can highlight it)is that the packaging – a pristine white box with a sleek, spotless aesthetic – is nicely branded with the Apple code. Make of it what you want.

The keyboard layout features six matte “agent” keys at the top of the keyboard that can be customized to perform specific tasks within ChatGPT or its Codex agent coding tool. Below are the six command keys that you can exploit to control these programs. You can pair your Micro device with your computer using Bluetooth or a USB cable.

Perhaps the most convenient thing the Micro offers is the voice dictation button – which means you simply tell the app what you want to do and it will do the work on your behalf. Just hold down the dictation button and start speaking. Once you’re done, click the “submit” button next to it to submit your request.

You can customize your Micro keyboard within ChatGPT itself, where the Micro tab allows you to adjust everything from the brightness of the key featherlight to the specific commands and designs you want to associate with those keys.

Die-hard developers – the device’s actual target audience – haven’t quite caught on to this. Redditor reviews have been largely negative, with one Reddit user calling it “joke not a real product” and others say that earnest developers won’t touch it. A review by the smaller, independent store Aftermath was even more sharplystating that the $230 price tag is tough to justify when compared to cheaper DIY products and off-the-shelf alternatives. (Title of this review: “OpenAI’s expensive macropad seems designed to piss me off on purpose.”)

It’s definitely the case that fresh users may need some time to learn how the Micro works and what to do with it.

Once I figured out how to program the keyboard to my liking, I found it to be really fun. You can assign different ChatGPT sessions to specific keys, allowing you to easily switch between all the different projects. If you combine this with a dictation button, the whole experience becomes much more capable and enjoyable.

But there is still a learning curve. Micro buttons are color coded. White means the agent is idle, blue means it is thinking, green means the task is complete, and red means an error has occurred. You need to remember this and also remember which specific designs are encoded for each key.

The large question is whether the Micro Keyboard is functionally easier to exploit than regular laptop work. In compact: why would I spend a week learning how to program and exploit this device when I already know how to exploit my computer’s mouse and keyboard?

Ultimately, your experience with Micro will depend largely on how often you exploit ChatGPT. Since I don’t exploit AI very often on a daily basis, I’m probably not its target audience.

That said, if you’re an advanced ChatGPT user, have $230 to spare, and like vintage hardware with clicky buttons, the Micro probably isn’t the worst purchase you can make – it might even brighten your day a little.

On Thursday, OpenAI said it had updated its ChatGPT desktop app to add support for ChatGPT Voice, allowing users to communicate with the app to control AI agents and perform tasks on their desktop.

The up-to-date feature leverages a up-to-date family of OpenAI voice models called ChatGPT-Live, which was released earlier this month.

OpenAI says ChatGPT Voice works with both ChatGPT Work and Codex, and can also exploit computer skills to search websites and applications. Additionally, on macOS, with Appshots, users can allow an app to access screen content, including alt text.

At launch, the smartphone version of ChatGPT Voice offered smoother conversations and better noise handling, but was not designed to take action on smartphones.

Thursday’s desktop update is more competent, allowing users to dictate elaborate commands involving multiple steps and respond when ChatGPT needs their input.

In a demo video, OpenAI showed a developer asking ChatGPT to create a up-to-date thread, submit a pull request, and find the root cause of the error with a single command.

ChatGPT Voice is now available on the desktop app.

Control your computer and direct multiple agents running in ChatGPT Work or Codex using only your voice.

It works on the GPT-Live platform, so it can simultaneously speak, listen and coordinate work in the application.

Rolling out today around the world… pic.twitter.com/ODZWKqecCf

— OpenAI (@OpenAI) July 23, 2026

The company said that users can exploit ChatGPT Voice on Codex from the iOS app via remote access.

Anthropic has also updated its voice mode for Claude, which can exploit Opus, Sonnet, and Haiku models to perform tasks in apps like Gmail, Calendar, Slack, Notion, and Canva.

OpenAI today announced that it is making ChatGPT Health, a feature that helps users with health-related questions, available to all U.S.-based users over the age of 18 on all plans. The announcement comes a day after a Florida pastor sued the company over donations an almost fatal suggestion not to consult a doctor.

The company began testing the feature through a dedicated hub in early January, allowing users to integrate data from other services and personal data from services like Apple Health, Function and MyFitnessPal.

According to OpenAI, users asked 230 million health-related queries per week. This number has already increased to 300 million.

Users can also integrate their medical records from hospital systems such as Epic and Oracle Health, and health platforms such as One Medical and Function Health.

OpenAI has previously said that users must apply the fitness hub for health-related queries. They can now draw conclusions from the combined information in the Health section in all queries. The company said that during testing it observed that 70% of health-related queries occurred outside of the dedicated center. With this up-to-date feature, users can apply their health information to obtain information about food or allergies in general chat.

The company noted that its models have made progress in answering health-related questions. The company noted that the smallest model in the latest release, GPT 5.6-Luna, outperforms GPT 5.5 in HealthBench evaluation, an open source benchmark developed by the company to evaluate large language models (LLM) for health queries. OpenAI says it doesn’t apply any user data to train its model and works with doctors to improve its models for health queries.

Despite the performance gains, the company maintains that its services are “not intended to diagnose or treat any medical condition.” The company cited these clauses in response to the above lawsuit and also informed about it New York Times that it is working to make health or medical related responses safer. In its latest implementation, OpenAI said it wants people to verify information and make medical decisions based on professional advice.

Several studies emphasized that AI bots are not reliable for medical advice. However, this hasn’t deterred companies like Anthropic and Google from launching health-related AI features.

ChatGPT Health will be available this week to signed-in users in the US on Free, Go, Plus and Pro plans on the web and iOS.

You may have heard that OpenAI released its first hardware this week: a $230 mini keyboard. You may not have heard that in addition to this “agent work command center”, OpenAI has also released ChatGPT basketball.

“This basketball is from the Pause. Play. Prompt. campaign, which aims to provide a physical reminder that creativity doesn’t just live on our screens,” the product listing explains. I couldn’t find any other mention of “Pause. Play. Hint.” campaign on the OpenAI website, but I understand this is OpenAI’s way of telling people not to spend all day on Codex.

Who says tech companies don’t think about our mental health?

Basketball will cost $70, or approximately 56 million input tokens for GPT-5. It is a 100% rubber ball that, due to its weather resistance, is better suited for outdoor play than the more pricey leather balls found on professional basketball courts. I’m glad to know that OpenAI sees a world where outdoor sports are possible, even as the boom in generative artificial intelligence accelerates greenhouse gas emissions from tech companies.

It’s difficult to imagine ChatGPT’s target basketball customer. Who is this for?

Step outside the safety of Silicon Valley filled with artificial intelligence and tokenmaxx, and you might worry that you’ll be intimidated for bringing a ChatGPT basketball court. You wouldn’t pay me $70 to get on the court in Philadelphia with this ChatGPT basketball. (If this were a free conference giveaway, it might be considered ironic – I really appreciate my “#FACEBOOK” bag, which is polished like A bar mitzvah party from the early 21st century.)

In defense of ChatGPT basketball, the AI ​​industry isn’t exactly known for its product-market fit instinct. May Humane Ai Pin rest in peace.

In addition to this $70 artifact, OpenAI also sells a line of merchandise with inspiring reminders such as “Good research takes time,” which I think is the perfect outfit for a startup founder meeting with investors demanding faster growth. There is also a certain amount of $175 zipper it says “research” in italics. The product description states that it “has a statement collar that’s reminiscent of when we were at university,” which may turn off the “I never went to college because I’m a coding major” crowd. (Also, can the subject reminisce about your days in academia? Should I expect grammatically correct sentences from people writing emails using ChatGPT?)

However, there is nothing wrong with a bit of company prestige. If OpenAI wants to commission any ceramic artists to honor the company’s history through functional tableware, I’d like to throw my hat into the ring.

According to reports, the first OpenAI solution on hardware devices is a mobile clever speaker with integrated AI capabilities that can sync with ChatGPT and provide other home AI services.

Bloomberg reported on Tuesday that the device – which is still in development – is expected to be screenless and is internally presented as a “human-like AI companion living at home.”

OpenAI has long said it wants to release a hardware product – there are rumors that it wants to release its own phone, which would put it in competition with Apple.

The newly released OpenAI device sounds like a departure from classic clever speakers — as sources described to Bloomberg a device that has a “personality” and, over time, is capable of actively learning about its owner, providing a more personalized experience. Sources say the machine would have access to the user’s digital life and download e-mails, for example.

The device is also oddly described as containing “mechanical components that can move on their own,” and the Bloomberg report details that the device is designed to “give the impression of a companion and become a physical manifestation of OpenAI’s ChatGPT.”

The device was developed with the lend a hand of many former Apple engineers who played key roles in “creating products like the iPhone and Mac,” Bloomberg writes. Indeed, OpenAI may be trying to release a recent line of hardware, but the company is currently up to its eyeballs in trouble due to hardware-related legal issues.

Last week, Apple sued OpenAI, accusing the artificial intelligence company of stealing trade secrets. Apple further argued that the allegations in the lawsuit are merely allegations the “tip of the iceberg” and that more abuses will be uncovered during the legal disclosure process. OpenAI denies any wrongdoing.

Citing anonymous sources with knowledge of OpenAI’s plans, Bloomberg writes that the company believes its recent product is “significantly different from anything Apple currently has on the market” and is “unlikely to infringe trade secrets” belonging to Apple.

OpenAI’s development comes as the tech world becomes increasingly excited about consumer AI hardware. Hark, the artificial intelligence lab founded by Brett Adcock, in May raised an oversubscribed $700 million Series A at a $6 billion valuation to build what it calls “personal intelligence” – proprietary artificial intelligence models combined with custom hardware designed to be a “universal interface between humans and machines.”

The company has yet to detail the shape of its device, emphasizing how much capital is chasing the category even before products ship.

More than three years after its launch, ChatGPT brought generative AI into the mainstream, OpenAI is expanding its focus not just to individual users but to families.

OpenAI is there hiring a dedicated product manager in San Francisco tasked with creating experiences for families, caregivers and seniors within its products. According to the job posting, the position requires products that build experience for parents and families and other trust-sensitive consumer experiences.

The hiring comes as ChatGPT’s audience expands beyond younger users. According to Sensor Tower estimates shared exclusively with TechCrunch, the share of ChatGPT users aged 35 and older globally rose to 31% in the second quarter from 26% a year earlier, while the share of users aged 18 to 24 fell to 29% from 34%. In the U.S., nearly one in four parent smartphone users used ChatGPT this quarter, up from 16% a year earlier, the company estimates.

OpenAI did not respond to requests for comment on the job posting.

The dedicated product role focused on families signals that OpenAI is starting to think of its products less as tools for individual productivity and more as technology aimed at households, said Ben Bajarin, chief executive of consultancy Imaginative Strategies.

“This is similar to the path that Google, Apple and Meta ultimately took as their platforms became embedded in everyday life, but AI ups the ante because the assistant doesn’t just mediate content or devices,” he told TechCrunch.

This change also brings novel challenges in terms of trust and security. Stephen Balkam, chief executive of the Family Online Safety Institute, said the hiring reflects both the maturation of OpenAI and the growing recognition that AI products used by children and teenagers require different safeguards than those intended for adults.

“I see it as security through redesign,” Balkam told TechCrunch. “You take the initial product or service that was released… not necessarily with kids in mind… so there’s a very necessary backlash and backlash.”

The comments come from novel research released this week by the Family Online Safety Institute found that parents underestimate how often their children utilize generative AI. While 27% of U.S. parents said their child used generative AI in the past week, according to a survey of more than 4,000 families in the United States and Australia, 38% of children said they did so on their own.

Balkam told TechCrunch that AI companies should build products differently for younger users, with stronger content controls, age-appropriate experiences, parental supervision and reminders that let users know they are interacting with AI rather than a human.

The hiring also comes amid growing scrutiny of how AI companies protect younger users. OpenAI has faced multiple lawsuits from the site parents accuse this ChatGPT contributed to the damage incurred by their children, including cases involving suicide.

Balkam said artificial intelligence companies have a chance to avoid the mistakes made by social media platforms, which treated children similarly to adults for years before adding stronger safeguards in the face of growing public pressure and regulatory scrutiny.

The hiring also fits into OpenAI’s broader efforts around families. During a recent workshop organized with the San Antonio Spurs Community Impact and Positive Coaching Alliance, the company he said its aim was to explore the role of artificial intelligence in youth learning, coaching and engagement.

That said, demographic changes are not unique to ChatGPT, although OpenAI’s audience is changing in some respects.

Sensor Tower estimates that users aged 25 to 34 make up 40% of the global audience for Anthropic’s Claude and Google’s Gemini, which equates to ChatGPT, compared to 33% for Microsoft’s Copilot. Copilot, however, skews older users, with 20% of users aged 45 and over, compared to 14% for Claude, 12% for Gemini, and 11% for ChatGPT.

While ChatGPT remains relatively underpenetrated among older users, it is adding them faster than competitors. According to Sensor Tower, the share of users aged 45 and over increased by three percentage points year-over-year in the second quarter, compared with a two-point raise for Copilot and a decline for Claude and Gemini.

Among U.S. smartphone users who are parents, Gemini had the broadest reach at 32% in the second quarter, followed by ChatGPT at 24%, Claude at 4% and Copilot at 2%.

For Bajarin, OpenAI’s decision to hire a family-focused product manager signals where consumer AI is heading. As AI becomes a shared technology across generations, he expects companies to introduce family plans, child and teen profiles, caregiver tools, shared home storage, AI learning and stronger security controls.

The Up-to-date York Times and The Daily News claim that OpenAI lied about its ability to search customers’ call logs and training datasets for their copyrighted works. It’s the latest escalation in a two-year lawsuit against the artificial intelligence company for allegedly violating copyright law by training its generative artificial intelligence models on Times content and reproducing that journalism in user results.

Throughout the case, OpenAI maintained that it had no way of searching its own training corpus. She also argued that searching or creating ChatGPT’s immense collection of conversations would be technically burdensome and raise user privacy concerns because the logs would need to be downloaded, processed and de-identified. Outlets sought this data to determine whether their copyrighted journalism was present in the OpenAI training dataset and whether and how often ChatGPT generated responses by using or reproducing their content.

In court-mandated April testimony, OpenAI privacy engineer Vinnie Monaco allegedly disclosed that OpenAI had already conducted internal searches and assessments of its training corps for copyrighted journalistic works.

Monaco’s testimony also allegedly revealed that, even before the NYT filed its lawsuit, OpenAI had already amassed a database of approximately 78 million de-identified ChatGPT conversations, which it used internally to determine the extent to which it was infringing the works of others. In addition to this dataset, OpenAI also allegedly implemented a “Bloom” filter as part of a toolkit called “Project Giraffe” that detected and recorded regurgitation in the results shortly after the lawsuit was filed.

These last two revelations are particularly significant. Plaintiffs originally asked OpenAI to provide a sample of 120 million chat logs, but OpenAI negotiated to reduce the sample to just 20 million. OpenAI eventually submitted this sample to the court last December, but it allegedly contained so many redactions that, in the court’s words, the sample became “useless.” The plaintiffs also alleged that OpenAI deleted billions of ChatGPT results after filing the lawsuit in direct violation of the court’s injunction, and that the AI ​​giant substituted millions of logs in the requested sample.

In other words, they argue that OpenAI made it unnecessarily hard to obtain information the company had already collected.

“If OpenAI truly believed that copying our clients’ journalism was fair and legal, it would not conceal the truth that it was doing so,” Ian B. Crosby, lead counsel for the plaintiffs, said in a statement.

Now the NYT and The Daily News are asking a judge to punish OpenAI for allegedly withholding evidence and interfering with the discovery process. They are asking the court to prevent OpenAI from using its 20 million chat log sample as evidence, saying it is unreliable; accept as a fact that the ChatGPT logs would show a earnest lack of closure and grounding of the plaintiffs’ content; to prevent OpenAI from arguing that the provided chat logs do not show significant regurgitation; and making OpenAI pay legal fees for having to pursue this evidence.

In a statement, OpenAI spokesman Drew Pusateri denied the allegations, accusing the Times of trying to gain access to users’ private conversations as the case faded away.

“As the Los Angeles Times’ case weakens and it is forced to drop its claims against us, it continues its efforts to violate the privacy of people who have nothing to do with this case, including by making blatantly false allegations,” Pusateri said. “We will continue to defend the privacy of our users and the long-established principles of fair use.”

OpenAI on Thursday unveiled its latest family of models, bringing a novel set of advanced programs to an increasingly crowded field of AI offerings.

GPT-5.6 comes in three flavors: Sol (considered a workhorse), Terra (a more in-between option), and Luna (a budget-friendly option). These models expand users’ capabilities in a variety of fields – the company promises powerful capabilities for enterprise work, coding, and even scientific research.

CEO Sam Altman recently promised that his company’s latest models are orders of magnitude more effective and cheaper than previous versions says CNBC that Sol is a 54% more effective token when it comes to AI coding tasks.

Most importantly, the company calls version 5.6 its “strongest cybersecurity model yet, achieving pioneering performance with significantly fewer tokens.”

Indeed, there has been much buzz about the model’s cyber capabilities, as the Trump administration has previously sought to limit its deployment, ostensibly out of concern that the model could be abused. GPT-5.6 supports defensive activities including threat modeling, code reviewing and patching, and blue teaming (simulating an attack on your own systems to find vulnerabilities before real hackers do).

OpenAI has also released a novel tool called ChatGPT Workwhich appears to have been designed as a workplace companion for corporate teams, running across desktops, web and mobile devices, that can facilitate with everyday office tasks such as creating documents, spreadsheets and presentations.

The newly announced OpenAI family of models follows similar releases this week from competitors SpaceXAI and Meta.

However, GPT-5.6 and its accompanying marketing seem best designed to target OpenAI’s main adversary, Anthropic. Anthropic has managed to gain traction in an underdog position in the AI ​​race by focusing on enterprise customers and, as a result, gaining an increasing share of support.

Not to be outdone, OpenAI quotes: Artificial analysis coding agent indexwhich is a noteworthy benchmark, and claims that its latest family of models outshines the Anthropic models in every respect.

OpenAI calls Sol its “best encoding model yet” and has pointedly compared it to Anthropic’s recently released (and very popular) Fable game. Using the Coding Agent Index, OpenAI claims that Sol “establishes a new state of the art of 80, 2.8 points above Fable 5, while using less than half the output tokens, taking less than half the time, and costing about a third less.”

He adds: “This advantage extends across the family: Terra performs slightly better against Fable 5, and Luna outperforms Opus 4.8.”

The company says that version 5.6 is now available on ChatGPT, Codex and OpenAI API. Availability per million tokens is priced as follows: Sol is $5 in/$30 out, Terra is $2.50 in/$15 out, and Luna is $1 in/$6 out.

OpenAI today released novel conversational models called GPT-Live-1 and GPT-Live-1 mini, claiming they sound more natural and are better at making turns. These are full duplex models, which means they can speak and listen at the same time, allowing users to interrupt naturally and providing features such as live translation.

The company is also replacing its current advanced voice mode in ChatGPT with GPT-Live-1 mini by default. Paid tier users will be able to access the larger GPT-Live-1 model. The previous model combined a speech-to-text model for transcribing speech, a enormous language model for generating responses, and a text-to-speech model for delivering the final response.

The company said at a press conference that the novel models address issues such as users being interrupted while talking and not having enough intelligence to answer questions. The novel OpenAI models will query the latest text-based models, such as GPT-5.5, for search, inference, or agenting capabilities while continuing the conversation.

OpenAI also showed that the model can remain hushed for a long time and absorb the context of the conversation until it is triggered. Additionally, because the novel voice mode has access to newer GPT models, it can also present certain information in a visual format. Other startups like Monogram, which raised $40 million in seed funding from DST and Lux ​​Capitalthey also rely on visual responses to boost the interactivity of assistants.

The company says the novel voice mode in ChatGPT is designed for longer conversations. During the briefing, ChatGPT Voice product manager Atty Eleti said he had 30 to 40 minutes of conversations using the voice feature during his walks.

OpenAI believes that voice can be the primary computing interface for performing elaborate work. Reports suggest that it may launch a pair of earbuds with AI features this year. However, no information has been provided about the hardware products.

“Over time, we believe this will also unlock the ability to use voice as the primary interface for processing data and managing the increasingly complex, long-term work of agents. We believe that voice can become the future interface for all kinds of work,” Eleti said.

Over the past few years, OpenAI has been working on improving voice features to make ChatGPT’s voice mode sound more natural. The company said that more than 150 million people talk to ChatGPT using features such as voice and dictation.

Rivals are also trying to make assistants more expressive.

Both Apple and Amazon have updated their assistants to provide more conversation and better context support. Startups like Sesame, founded by Oculus co-founder Brendan Iribe and Ankit Kumar, have also launched AI assistants that enable more natural conversation while performing background tasks.

OpenAI is moving in the same direction, aiming to allow users to talk to their assistant for longer periods of time without using their hands. Despite claims that the novel voice mode sounds more natural, the company stressed that it has no intention of making it an AI companion. It noted that the novel models have built-in safeguards to ensure teens receive age-appropriate responses and provide resources if the conversation turns to topics such as self-harm.

The novel voice mode still needs work. During a demonstration when the company showed off its live Hindi translation feature, the assistant had a bulky American accent and spoke in Hindi that sounded unnatural and had a slightly bookish tone. The company said the novel mode is optimized for “most languages ​​in use,” but did not specify which ones.

SpaceXAI has released its latest Grok 4.5 model – the first since the company went public a few weeks ago.

IN blog post released Wednesday, SpaceXAI characterized its modern version as a workhorse that can handle all the common tasks that the AI ​​industry has sought to automate: coding and application development, office and clerical work, research, writing and other forms of routine knowledge work.

Grok can reportedly do all this for less money, as SpaceXAI claims its model has “twice the token yield” of other leading models. If this translates to real-world exploit cases, this efficiency will be a gigantic advantage for SpaceXAI as the cost of tokens becomes an increasing concern for AI consumers.

On Wednesday, the company released benchmarks that appeared to show the Grok’s competitiveness relative to other top-of-the-line models from SpaceXAI’s competitors, though not far from best-in-class:

IN post on his social media platform X (a subsidiary of SpaceXAI), founder Elon Musk compared the model to Opus, Anthropic’s LLM designed for intense and convoluted tasks.

“Based on the very positive feedback from customers participating in our beta testing program, @SpaceXAI will release Grok 4.5 to the public tomorrow. It is an Opus-class model, but faster, more efficient token and cheaper,” Musk wrote in his X post.

Musk later added: “Our internal assessment is that Grok 4.5 is roughly comparable to Opus 4.7, but significantly faster. The combination of capabilities, greater speed and lower cost makes it competitive.”

SpaceXAI says its modern model costs $2 per million input tokens and $6 per million output tokens. That’s quite competitive if Grok’s capabilities match SpaceXAI’s rhetoric.

Opus 4.7for comparison, it costs $5 per million input tokens and $25 per million output tokens. OpenAI has tiered costs for different versions of the model: Sol, the most pricey, costs $5 for 1 million input tokens and $30 for 1 million output tokens, while the cheapest Luna costs $1 for 1 million input and $6 for 1 million output tokens.

This is an vital week for AI model launches. OpenAI is there planning to release GPT 5.6, its newest and most powerful model, on Thursday. The launch of this model was previously confined by the Trump administration due to concerns about security consequences. OpenAI he called it is the “strongest model yet.”