OpenAI puts Deep Research inside Codex and ChatGPT Work
Cited, multi-source research now runs from your coding agent — and OpenAI quietly folds Voice's Instant/Medium/High tiers into plain reasoning effort.
Copy markdown
Your coding agent can now do the reading
Deep Research now runs inside Codex: pose a complex question and it works across the web, your files, and connected apps, then returns an editable, cited document. Your agent can scout an API or nail down a spec before it writes a line — no switching over to ChatGPT.
It reaches ChatGPT Work too, with real sharing
The same Deep Research lands in ChatGPT Work on web, desktop, iOS and Android for Plus, Pro, Business, Enterprise and Edu users. Alongside it, the Library now shares files and folders with Viewer/Editor roles — handy when a team is passing prompts and assets around.
Voice tiers collapse into reasoning effort
OpenAI deprecated the separate Instant/Medium/High Voice intelligence levels. You now pick a model — GPT-5.6 or GPT-6 Astra — and set reasoning effort instead. If you scripted flows or habits around those tiers, that lever is gone as of this update.
New GPT-Live ceilings, by plan
Live-voice limits got simpler: Go and Plus get up to 3 hours a day on GPT-Live-1, Pro at $100/mo gets 15 hours, and Pro at $200/mo is unlimited. Worth re-checking if you lean on live voice for real work.
Elsewhere: Google's agent sandboxes hit GA
On Google's Gemini Enterprise Agent Platform, Computer Use and Shell sandboxes reached general availability — isolated command execution with pause/resume, CMEK and VPC Service Controls. A signal that browser-and-terminal agents are now production-grade if you build on Google Cloud.