OpenAI DevDay leaks: an always-on "O" agent, a $500 Pro Max tier, and GPT-6 Cyber
DevDay is Tuesday, but the announcements are leaking early: an always-on agent called "O," a $500 Pro Max tier, a GPT-6 Cyber preview, and new API platform plans.
OpenAI's DevDay keynote is Tuesday, September 29 at Fort Mason in San Francisco, with Sam Altman opening. But the announcements are leaking early, and this year's theme is already clear: agents that never sleep, and pricing for people who want them to run fast.
Here is what has surfaced so far. None of it is confirmed by OpenAI, so treat this as the rumor roundup it is.
An always-on agent called "O"
The most interesting leak comes from TestingCatalog, which found references to an agent called "O" in ChatGPT's configuration: "O" as a display name and "-o" as an email suffix. That suffix suggests the agent gets its own email identity out of the gate. The name also showed up briefly on the upgrade page for the $100 Pro plan, and OpenAI's developer account posted what looks like a teaser for it today.
The idea is a persistent agent that keeps working outside a normal chat session: recurring research, monitoring, communications, the kind of work you want handled while you are away. TestingCatalog notes it may connect to "Aeon," an internal name previously tied to OpenAI's always-on agent work and custom agents for ChatGPT Workspace accounts. Meta has been pushing this category forward with Muse, which adds pressure on OpenAI to show its own persistent agent, and DevDay is the obvious stage.
A $500 "Pro Max" tier
Independent researcher Tibor Blaho dug references to a "ChatGPT Pro Max" plan out of hidden interface logs. TestingCatalog and the leaks account Chetaslua confirmed it with screenshots of the tier sitting inside ChatGPT's own subscription UI. The price: $500 a month, with some screenshots showing $600 after local taxes. That is 25 times the $20 Plus plan and more than double the $200 Pro tier.
The pitch, per the leaked strings: "Fastest Work and Codex," meaning top speed on Work (multi-step research and writing) and Codex (long-running autonomous coding), plus the most capable frontier models, maximum memory, 100GB of file storage, and early access to new models. That last perk is awkward, since the $200 Pro plan already promises early access.
Context matters. OpenAI paused new sign-ups and upgrades to the $200 Pro plan on September 10, and power users have been asking for more headroom. The speed story may trace back to Cerebras: OpenAI has been deploying Cerebras wafer-scale hardware, launched Codex-Spark on it at over 1,000 tokens per second earlier this year, and demoed GPT-5.6 Sol at up to 750 output tokens per second in an Ultrafast mode. Neither company has confirmed the link.
For pricing context, Anthropic's Claude Max runs $100 to $200 a month, and Google cut its AI Ultra plan from $250 to $100 back in May. A $500 tier would put OpenAI in a pricing league of its own.
GPT-6 Cyber
Fortune reported that OpenAI is about to preview GPT-6 Cyber, a cybersecurity-focused model, with DevDay as the likely unveiling. A limited group in OpenAI's application-only Daybreak Red program already has alpha access. The report also claims OpenAI is building a product to help customers deploy AI more securely, and that the company plans to ship "a dozen or more" products around DevDay.
This would be the fourth dedicated cyber release this year, after GPT-5.4-Cyber in April, GPT-5.5-Cyber in the summer, and GPT-5.6-Cyber in August. It lands weeks after OpenAI acknowledged its flagship model can at times try to evade human oversight, which makes the timing interesting.
New API platform plans
TestingCatalog also spotted a new onboarding flow for OpenAI's API platform with three plans: Free (experimenting, with free API access through Codex and limited Playgrounds), Prototype (lightweight apps, all models, higher usage), and Accelerate (production, starting at $50, all models, higher rate limits). The framing suggests OpenAI wants the platform to become a full development environment rather than just an API dashboard, which fits the Agents API direction from earlier this month.
What to watch Tuesday
DevDay opens Tuesday morning at Fort Mason. If even half of this ships, it is a statement that OpenAI's developer story is now agents-first: persistent agents, faster agents, and a platform to build them on. I will update this post as the keynote lands.
Sources: TestingCatalog, Fortune via EM360, BleepingComputer, Decrypt. All items above are unconfirmed leaks, not official announcements.
