OpenAI’s move into hardware arrived on 15 July in a form almost nobody had reported: a small mechanical keypad for developers, built with an accessory maker, sold in limited numbers at $230 (about £172). It landed a day after Bloomberg described the product most coverage has since treated as OpenAI’s first device — a portable, screen-free home speaker with a camera and parts that move on their own — and five days after Apple asked a federal court to stop OpenAI using what it says are stolen trade secrets.

The distance between those two products is the story. One is real, priced and shipping this month, and does a narrow job for people who already pay for Codex. The other has never been announced by the company said to be building it, carries no confirmed specification, no confirmed price and no confirmed date, and sits behind an injunction request that could move whatever date exists. For marketing teams being told to prepare for ambient, screenless AI, only one of those is a thing that can be planned around.

What has OpenAI actually shipped?

Codex Micro is OpenAI’s first hardware product to reach customers, released on 15 July at $230 (about £172) and co-designed with Work Louder, a Canadian-Italian keyboard maker. The unit carries 13 mechanical switches, a touch sensor, an analogue stick and a rotary dial that sets how much reasoning effort an agent spends on a task. Its agent keys light up with the live status of each Codex session, so a developer can see what is thinking, running or finished without switching windows. It is configured through the ChatGPT desktop app, and comes in clicky and silent versions. Pre-orders opened at launch with first units due to ship this month.

OpenAI told TechCrunch the Micro is a limited-run collaboration, which places it closer to a signal than a product line. The audience is real but specific: Codex and ChatGPT Work have reached 8 million active users, according to Codex engineering lead Thibault Sottiaux, and OpenAI merged Codex into the ChatGPT desktop app the week before the keypad appeared.

The design idea inside it travels further than the object. Codex Micro takes controls that normally live in a menu or a chat box and puts them under the fingers, with a physical dial for the trade-off between speed and thoroughness. That is a legible answer to a problem every team running AI agents has: the controls that matter are buried in the interface, and nobody adjusts what they cannot find.

How much of the speaker has OpenAI confirmed?

OpenAI has confirmed nothing about the speaker. Every detail comes from Bloomberg’s Mark Gurman, reporting on 14 July from people who asked not to be identified because the project has not been announced. On that account the device is a mobile, screen-free smart speaker pitched internally as a new type of home computer for the AI era and a humanlike AI companion that lives in the home. It has a camera and sensors to read its surroundings, a rechargeable battery so it can be carried between rooms, mechanical elements that move on their own, and GPT-Live, the full-duplex voice model OpenAI released on 8 July, doing the talking. Bloomberg’s sources put the reveal later this year, the sale in 2027 and the price between $200 and $300 (roughly £150 to £225).

The market took the report seriously even without confirmation. Shares in Sonos fell more than 10% in late trading on the day before paring the losses, Bloomberg reported, while Apple fell less than 1%.

One distinction is worth holding. At the GPT-Live press briefing on 8 July, OpenAI told reporters it was not aiming to make the voice model an AI companion, and published expanded safety testing covering emotional reliance on AI. Six days later, unnamed sources described the device running that model as a humanlike companion whose defining feature is the relationship it forms with its owner. Both statements can stand at once, because a voice model is not a product. The gap between them is the gap between what OpenAI says on the record and what its staff say when they are not named.

The one figure OpenAI has published is the useful one: more than 150 million people use ChatGPT’s voice and dictation features every week. That is evidence an audience for voice exists. It is a number about software people already own, not hardware anyone has been asked to buy.

Could Apple’s lawsuit move the date?

Apple sued OpenAI, io Products and two former employees on 10 July in the US District Court for the Northern District of California, alleging that the theft of its hardware trade secrets ran, in the complaint’s words, at every level, from members of its Technical Staff to its Chief Hardware Officer. The named individuals are Tang Tan, OpenAI’s chief hardware officer and a 24-year Apple veteran, and Chang Liu, a former Apple engineer. Apple says more than 400 former Apple employees now work at OpenAI and has described the filed allegations as the tip of the iceberg. It is seeking damages and an injunction barring OpenAI from using the material.

OpenAI has denied wrongdoing, told Bloomberg it takes the allegations seriously but is not aware of any evidence the complaint has merit, and believes the device veers significantly from anything Apple has on the market today. The company paid $6.5bn last year for io Products, the hardware start-up co-founded by former Apple design chief Jony Ive, whose LoveFrom studio is working on the wider range. OpenAI is also expected to go public within months, which makes an unresolved injunction request a disclosure problem as well as an engineering one.

Codex Micro sits outside all of that. A keypad assembled with a specialist accessory maker touches none of the industrial design, supplier relationships or metal finishing the Apple complaint describes, which is one reason it could ship in the same week the lawsuit was filed.

OpenAI has not said when it will unveil the speaker, what it will cost or what it will be called. The only hardware the company is currently taking money for is a limited run of keypads for people who write software.