Leaked DevDay Details: OpenAI’s Rumored “O” Always-On Agent and Faster APIs

A set of leaks summarized by Geeky Gadgets and credited to World of AI lays out several product ideas and model updates that are reportedly surfacing around OpenAI's DevDay. The clearest consumer-facing item in those leaks is a feature labeled "O," described as an always-on assistant intended to take on long-running or complex tasks without constant human supervision.

The leak frames the "O" agent as a persistent assistant that can manage tasks such as coordinating workflows, organizing schedules, or handling email. It's said to be powered by a technology called Astra AON, support 63 languages, and enable multiple AI agents to collaborate on a single process. The pitch in the leak is automation for tasks that normally require continuous oversight or repeated user prompts.

Another thread in the leaks covers infrastructure: an Ultra-Fast API tier. The materials mention three speed tiers—Standard, Fast, and Ultra-Fast—with the Ultra-Fast mode reportedly able to process up to 750 tokens per second. The implication is lower latency and higher throughput for applications that require near-real-time interaction, such as streaming or interactive tools.

The leak pack does not focus only on OpenAI. It highlights Anthropic's Sonnet 5.5, which the leaks claim outperforms OpenAI's GPT-6 Soul on several benchmarks and offers faster performance and lower cost for some workloads. Miniax's M3.1 Flash is mentioned as a model tailored to coding and debugging tasks. These entries suggest the leaks are positioning multiple vendors' releases alongside OpenAI's announcements.

The leaks also include a preview of Longat 2.5, described as a privacy-forward multimodal model with a million-token context window and zero data retention. If accurate, that combination would target applications that need very long context and tighter data handling guarantees.

If the leaks are accurate, DevDay could include announcements that touch three practical developer concerns: agents that reduce manual oversight for recurring workflows, higher-throughput APIs for low-latency apps, and specialized or competitive models that change cost/performance calculations. The "O" agent concept points at automation that acts over time, the Ultra-Fast API speaks to responsiveness and scale, and the other models raise questions about price, latency, and task specialization.

Practical implications for users and teams

• For individual users: an always-on assistant that safely manages email or scheduling could reduce routine time sinks, but it raises questions about access controls and safety.

• For procurement and ops teams: competing models that claim better benchmarks or lower costs will affect vendor comparisons and total cost of ownership.