OpenAI Hardware · Codex Micro
OpenAI ships its first hardware.
One press, one agent.
OpenAI has shipped Codex Micro, the first-party hardware device it has ever built. Two physical buttons on the palm-sized body invoke either the coding agent Codex or the general assistant GPT-5.6 in a single press. It is the first shipping product to emerge from the io acquisition roughly two years ago, and the first time an AI-first cloud company puts a "press-to-summon" agent into the customer's hand.
The News
OpenAI ships
a press-to-agent device.
Two cloud-only years, then a first piece of first-party hardware.
In July, OpenAI announced Codex Micro, the first hardware product ever released under its own brand. The palm-sized body carries two physical buttons, an on-board microphone and a speaker: pressing one summons the code agent Codex, pressing the other summons the general assistant GPT-5.6. Pairing runs through the company's companion mobile app, and the device talks to OpenAI's backend over Wi-Fi and Bluetooth.
It is also the first shipping product OpenAI has released since Sam Altman acquired Jony Ive's io in 2025. For the past two years the company's line-up consisted of ChatGPT and Codex — cloud only, no hardware at all. OpenAI Newsroom frames the device as the company's first attempt to place its models "in the hand, not in the pocket".
By the Numbers
The device, in figures.
Priced against the 2024 wave — Rabbit R1 at $199, Humane Ai Pin at $699 plus a monthly plan — Codex Micro is landing squarely inside the category's standard range. The latency figure is a round-trip estimate derived from OpenAI's own Codex API benchmarks, and the eight-hour battery number is the category-standard baseline for an always-on ambient companion.
Why It Matters
Why hardware, why now.
Two years since io, the Rabbit and Humane lessons, and the ambient endpoint race.
The io acquisition, cashed in
OpenAI bought Jony Ive's design firm io in 2025 for roughly $6.5B and quietly stood up an internal hardware team. Codex Micro is the first artifact that team has shipped — the first tangible piece of OpenAI you can hold — and therefore the first real milestone for judging that acquisition's return.
The Rabbit and Humane lesson
The 2024 wave — Rabbit R1 and Humane Ai Pin — never controlled their own frontier model, and their UX stalled on latency and accuracy. OpenAI is entering with the one advantage those two lacked: a first-party GPT-5.6 and Codex backend it owns end-to-end.
Owning the ambient endpoint
Being able to summon an agent without unlocking a phone — while cooking, driving, or juggling a task at the desk — is the shortest path to inserting AI into the moments where phones cannot follow. This is the next battleground behind Meta Ray-Ban: the ambient endpoint that sits between the user and the model.
Who's Affected
Who feels it, and how.
Engineers
Alongside the desk-side Codex CLI, a "pocket Codex" for one-press code edits and commits becomes real. A fix that surfaces during a commute or a meeting can be spoken into the device and left to run back-of-desk, ready to review when you return.
Business buyers
Procurement costs a few hundred dollars per head, but per-device invocation counts turn productivity into something you can actually measure. A ten-unit, single-department PoC is a reasonable second-half experiment before any broader commitment.
Product managers
The design question becomes how to replicate the one-press UX in your own product. Entry points that once took a notification and three taps are now, seriously, on the table for a dedicated button or single-shot voice trigger.
The Counterpoint
Japan's operators
should sit this round out.
Three reasons not to buy on day one.
1) A Japan launch date and Japanese-language support are, as of today, unconfirmed. Deploying a device designed around an English-first UX into Japanese business workflows at this stage is premature. 2) The 2024 wave showed that AI hardware has a wide gap between opening interest and durable use; the OpenAI brand plus Jony Ive's industrial design is a strong hand, but not a guaranteed bridge across that gap. 3) An always-listening microphone reshapes how you handle work data and conversation privacy. Meeting-room and open-office use policies must be written before, not after, the first device lands.
What to Do Next
What to actually do.
| Short term (next 3 months) | Mid term (next 12 months) |
|---|---|
| Source one or two units through a US-bound teammate and stress-test the Codex button on real, day-to-day work. | Draft the departmental PoC proposal now, so it is ready the day OpenAI confirms a Japan launch and Japanese language support. |
| Get security and legal on the same page about always-on microphones before any device reaches an employee's desk. | Write the internal guideline for device × Codex operations — Git permissions, audit logs, rollback procedure — as a formal document. |
| Run one internal product exercise: prototype a "one-press" invocation in your own app to feel the UX debt. | Track the responses from Anthropic and Google every quarter; the 2027 v2 wave is where the market decision gets made. |
Codex Micro is the opening move of the "AI-delivered-as-hardware" era, not its final verdict. For Japanese operators, the rational play is to avoid rushing direct procurement while lining up the UX and operational scaffolding early. In parallel, Anthropic's and Google's answering products deserve quarterly review — this is now a category worth watching.