Tech • AI • Robotics • Game
OpenAI's New Agent O Changes ChatGPT Forever
9.3/10TL;DR
A leaked ChatGPT Pro subscription screen and backend configuration have fueled speculation that OpenAI is preparing an always-on assistant called O, as rivals including Microsoft, Google, Meta, xAI, and Anthropic push deeper into persistent AI agents.
KEY POINTS
Leak points to a paid always-on assistant
A widely shared screenshot posted on September 25 showed a $100 ChatGPT Pro plan labeled standard and described as Frontier Intelligence. Alongside model access, memory, 100 GB of storage, and early feature access, the page listed an always on assistant. The placement inside a subscription offer made the leak notable because it suggested a product positioned for paying users rather than an experimental internal tool.
The name appears to be O
Separate configuration details surfaced with the display name O and an email suffix linked to the same name. That raised the prospect of an assistant with a persistent identity that could remain reachable after a chat session ends. The evidence does not confirm that it can send email, read inboxes, or independently manage communications, but it points to a more durable assistant model than a standard chat interface.
Key details remain unconfirmed
No official announcement has established how O would work, what permissions it would receive, whether it could schedule tasks, how long it would retain memory, or who would get access. The leaked $100 tier also does not settle final pricing, regional availability, or whether a higher tier such as a rumored Pro Max plan exists. Claims of a Tuesday launch, support for 63 languages, or broad rollout across all pro users remain unverified.
Dev Day has become the focal point
Attention is centered on September 29, when OpenAI Dev Day is scheduled at Fort Mason in San Francisco with Sam Altman set to keynote at 10:00 a.m. Pacific. That timing has turned the leak into a plausible pre-event signal, but not a confirmed reveal. The distinction matters because anticipation around the event has already pushed speculation ahead of confirmed facts.
AON adds another layer of ambiguity
The leak intersects with earlier references to AON, a name previously tied to always-on agent efforts and custom agents for workspace accounts. One reading is that O could be the consumer-facing expression of similar infrastructure. Another is that AON and O are separate projects, internal labels, or different versions of the same system.
The strategic appeal is clear
An always-on assistant would be valuable if it could keep working on research, monitoring, and communication tasks without requiring users to reopen a conversation. The central test would be reliability: whether it can remember objectives, detect changes in circumstances, and know when human input is needed. Those are the practical features that would determine whether O is a breakthrough product or simply a new branding layer.
Competition is accelerating in persistent agents
Rival companies are moving in the same direction. Reports describe Meta launching Muse for tasks including email and payments, xAI introducing Grobbot, Anthropic developing scheduling features and deeper workplace integrations, and Google building long-running assistant experiences around Gemini. The common theme is a shift from one-off prompts toward agents that continue operating between interactions.
Microsoft has shown the clearest blueprint
Microsoft on September 25 outlined a rebuilt Copilot centered on Home, Code, and Autopilot. The company said Autopilot can watch channels, follow threads, run recurring work, and resume projects days later inside an organization’s tenant using its own identity, memory, computer, and workspace. That creates a concrete benchmark for what enterprise buyers may expect from any persistent assistant introduced by competitors.
Infrastructure and speed may also be part of the story
Separate testing findings suggest OpenAI may also expand compute options around the event, including a hidden selector for standard, fast, and ultrafast processing in the Responses API Playground. Reported figures for GPT-5.6 Sol reached up to 750 output tokens per second, with Cerebras linked to the fastest mode. Even if those are peak numbers rather than guaranteed real-world results, they indicate that infrastructure and latency could be a major part of the coming product strategy.
Google is taking agents in a more visual direction
Google has introduced Gemini 3.8 Live with a responsive avatar for enterprise use. It combines real-time speech, expressions, lip sync, and tool use across 97 languages, allowing background operations to run while the conversation continues. The contrast highlights how the market is branching into both invisible task agents and visible conversational personas.
CONCLUSION
The leak around O suggests OpenAI may be preparing a persistent assistant designed to outlast a single chat session, but the core product details are still unknown. With competitors already defining the category, the next confirmed move could shape how always-on AI agents are priced, governed, and used at scale.
Ask a question