OpenAI preps always‑on 'o' agent, faster API for DevDay
TestingCatalog says internal ChatGPT configuration points to a consumer assistant called 'o' and a Cerebras-powered speed tier, both expected around Tuesday's developer conference.
OpenAI is preparing to introduce an always-on assistant called "o" at its developer conference on Tuesday, TestingCatalog reported, citing references to the name inside ChatGPT's internal configuration.
https://x.com/testingcatalog/status/2103787365986623925
TestingCatalog said internal strings list "o" as a display name and "-o" as an email suffix, suggesting it will handle email on a user's behalf, and that references briefly appeared on some accounts' ChatGPT upgrade screens. Separately, the newsletter writer kimmonismus said on X that "o" would ship on all ChatGPT Pro tiers.
OpenAI is also expected to widen access to an "Ultrafast" inference tier around the same event, according to a separate TestingCatalog report. The outlet said the mode, built on Cerebras hardware, can push GPT-5.6 Sol to roughly 750 tokens a second. That is up to 14 times the speed of the standard tier, though it remains limited to select customers today.
The X account mark_k described "o" as comparable to Grok's always-on bot and Meta's Muse, saying it would offer "a computer in the cloud, which is always online," without citing a source for the claim.
Neither claim is confirmed. TestingCatalog said it could not determine which tasks, permissions or memory features "o" will support, or whether every upcoming GPT-6 model will work with Ultrafast at launch. OpenAI has not commented publicly on either report.