Pick Your Own AI Model in Workflows, and Tune How Hard It Thinks
The AI Agent Action inside your workflows just got a lot more flexible. Instead of being locked to one AI provider, you can now choose between three, pick the exact model you want, and even control how much thinking it puts into each task.
It sounds technical, but it's genuinely simple to use, and it can make the same automation run faster and cheaper. Let me walk you through it.
What landed
Your AI Agent Action now supports models from three providers, all in one place:
- Anthropic: Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5
- Google: Gemini 3.6 Flash, Gemini 3.1 Pro Preview
- OpenAI: GPT-5.6 Luna, GPT-5.6 Tera, GPT-5.6 Sol, GPT-5 Nano
You no longer have to guess which one suits the job. The picker groups models by provider and gives each a short description of what it's built for, so you can weigh your options before you commit.
The redesigned model picker
Open your AI Agent Action and click the model dropdown. In the screenshot below, you'll see the models laid out by provider, each with a plain-English note on what it does best.
Look for the thinking chip next to certain models. That marks the ones that can apply deeper reasoning, which matters when a task involves several steps or needs to be accurate rather than just fast.
Set how hard it thinks
On models that support thinking, you can also set the reasoning effort to Low, Medium or High, right from the same dropdown.
Higher effort means deeper reasoning for complex tasks. Lower effort returns faster answers for simple ones, and uses fewer tokens doing it. You can set this on each action independently, so you're never paying for more thinking than the task needs.
How to set it up
- Open your AI Agent Action.
- Click the model dropdown.
- Browse by provider, and use the thinking chip and model description to pick the right one.
- Choose an effort level from the same dropdown.
- Save and publish.
The practical upshot is simple. Use a flagship model for the multi-step work where accuracy really counts, and a lighter, faster model for the easy jobs like sorting intent or reading a simple field. Combined with the earlier token savings, the same task can now run up to 50% cheaper.
Have a workflow with an AI Agent Action already running? Open it up and take a minute to match the model to the job. It's a small change that can pay off every single time that automation fires.
