OpenAI and Google are about to release new models

Edition 135 | November 10, 2025

If your entire system is built on top of Chromium, maybe you shouldn’t advertise yourself as the “Chrome killer.”

Welcome back to Building AI Agents, your biweekly guide to everything new in the field of agentic AI!

In today’s issue…

• OpenAI and Google prepare dueling model releases

• AI agent toolkits for languages beyond Python

• A web search API built for agents

• Microsoft’s head of superintelligence on AI for humanity

…and more

🔍 SPOTLIGHT

OpenAI and Google are getting ready to round the next corner in the AI race.

Over the past few days, traces of a new model called “GPT-5.1 Thinking” have been appearing in the code behind the ChatGPT interface, indicating that this experimental option is occasionally being secretly served instead of the standard GPT-5 thinking mode. At the same time, select users of Google’s Vertex AI platform have been given the option of choosing a new “gemini-3-pro-preview-11-2025” in place of Google’s current state-of-the-art Gemini 2.5—openly, in this case.

Both kinds of testing are common ones for new model releases, and they’re not as different as they seem: OpenAI is perfectly aware that tech-savvy ChatGPT users are constantly looking into the client-side JS code that serves its web app, and there’s no way they’re naïve enough to think that the sneaky “gpt-5.1-thinking” snippet would stay hidden. Since it would be trivial to use a code name within the application to hide the new model’s identity, the fact that they haven’t done so indicates they want the world to know that GPT-5.1 is coming. It’s a method of building hype as much as actually testing the model.

So now that we’re taking the bait, how transformative should we expect these models to be? Is GPT-5.1 going to be the quantum leap that GPT-5 wasn't? Will Gemini 3 Pro be AGI?

Probably not. LLM progress has slowed significantly, making new drops feel more like the annual iPhone release than potential world-changers. That’s not to say it’s impossible that OpenAI or Google (or Anthropic, their only real peer in the AI race) will hit a huge breakthrough that will feel like GPT-3 or GPT-4 did—after all, they’re spending hundreds of billions of dollars to try to make it happen. But if they do, it probably won’t be named something evolutionary like GPT-5.1. That’s the kind of name you give to a model that pushes the frontier forward by feet, not miles.

Still, GPT-5 came out only 3 months ago, and even incremental changes can add up to incredible progress when they come so quickly after each other. The steady drumbeat of new releases is paying off: the length of tasks AI agents can perform is doubling every 7 months, an unfathomably fast rate when you extrapolate it over time.

AI agent builders should be prepared to test these new models in their current systems the moment they come out. Have a set of standard tasks, running from easy to hard, that you put each of your agents through, and record how well they perform. When a new frontier LLM gets released, plug it into your agents and compare. Is it better, faster, cheaper? Time to switch over.

Revolutions are often made of many small evolutions, and the newest evolution looks to be just around the corner.

Always keep learning and building!

—Michael