During a routine test on Codex yesterday, GPT-5.5 briefly appeared as the default model in the selector for all users, free and paid. It disappeared quickly, but not before people took screenshots. Along with it came a list of internal model names nobody had seen before: oai-2.1, arcanine, glacier-alpha, and several glacier-alpha variants, all with descriptions that raised more questions than they answered. Glacier-alpha is described as "intelligence that moves continents." Arcanine as a "frontier model with legendary appetite for starches." Nobody knows what that means. Everyone is talking about it.

GPT-5.5 showing up as the default is the obvious headline, but the list underneath it is the more interesting part. Heisenberg, a life sciences research model. GPT-Rosalind, which we covered last week. Multiple glacier variants. The sheer number of models in the pipeline suggests OpenAI is building out an entire family of specialized models simultaneously, not just iterating on one flagship.

Sam Altman has been dropping hints all week that something big is coming Thursday. Between the accidental model picker reveal and the Spud outputs that have been circulating, the signal is about as loud as it gets without an official announcement. Prediction markets have GPT-5.5 launching as soon as tomorrow.

We will find out soon enough.

OpenAI officially launched workspace agents today, and the pitch is straightforward: describe a job that needs doing, and ChatGPT builds an agent that does it continuously without you having to be there. These are not one-off assistants you prompt when you need help. They run on a schedule, pull context from your docs, emails, and chats, take actions inside your tools, and keep going in the background while you work on other things.

The templates give a clear picture of what OpenAI thinks businesses will actually use this for. There is Spark, a lead outreach agent that researches inbound leads, scores them, drafts follow-up emails, and updates your CRM. Tally pulls data every Friday, builds charts, writes a summary, and shares a report with the team automatically. Scout monitors Slack and support channels and turns feedback into prioritized tickets. Angle acts as a marketing strategist, creating production-ready web pages on demand.

In Slack specifically, agents can jump into threads, understand what is being asked, pull the right context, and resolve issues without anyone having to loop them in manually.

OpenAI is not just building a better chatbot. They are building the operating system for how teams get work done, with ChatGPT sitting at the center of it all. Whether that vision lands depends on how well the agents actually perform in real workflows. But the ambition is clear.

Google Cloud launched the Gemini Enterprise Agent Platform today, a new product designed to let businesses build, scale, and manage AI agents across their organization. The timing is hard to miss. OpenAI launched workspace agents this morning. Google answered by the afternoon.

Every major AI lab is now making the same bet: the next big product category is not a smarter chatbot, it is an AI that keeps working after you close the laptop. The race to own that layer of enterprise software is now officially on, and it is moving fast.