Models
Run managed Gemini models without an API key, or connect OpenRouter to pick custom models for AI Agent. Managed model usage is billed through Glow and remains subject to usage limits and model availability.
- Use it to
- Compare managed Gemini models (Gemini 4 Argon default, Flash, and Pro) and custom OpenRouter models for Glow's AI Steps
- Find it
- AI Agent Step → Setup tab → Brain (model) dropdown (AI Prompt and AI Data Transform use managed Gemini defaults)
- Cost
- Managed Gemini models use Glow AI tokens · OpenRouter models bill model usage to your OpenRouter account plus Glow Step credits
- Limits
- Model selection in the Brain dropdown is available on AI Agent only
- Instead
- External AI provider connectors in Dock → Apps → AI (to call OpenAI, Anthropic, or Mistral actions directly with your own API key)
AI Agent also supports models through a workspace OpenRouter connection, with model usage billed to that account.
Which step chooses a model
Only one AI step lets you pick a model: AI Agent.
| Step | Model selection |
|---|---|
| AI Prompt | Managed default Gemini model for generated answers, with optional shape repair. |
| AI Data Transform | Managed default Gemini model, extracting ordered values across parallel calls. |
| AI Agent | Brain (model) dropdown in the App drawer. Select managed Gemini models or connected OpenRouter. |
If you want control over the model, use the AI Agent step.
Selectable models in the Brain dropdown
When configuring an AI Agent, the Brain (model) dropdown in the App drawer offers two sources:
Glow managed Gemini models
Configure the agent’s instructions, tools and output before testing:
- Gemini 4 Argon (Default): The default selection for new AI Agent Steps.
- Flash models: Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash, Gemini 3.5 Flash Lite, Gemini 3.1 Flash Lite, Gemini 3 Flash Preview, Gemini 2.5 Flash, and Gemini 2.5 Flash Lite.
- Pro models: Gemini 3.1 Pro Preview and Gemini 2.5 Pro.
Custom models (OpenRouter)
Switch the AI Agent’s provider to OpenRouter. Workspace administrators can connect or reconnect OpenRouter inline from the setup panel (or under Workspace Connections), then choose any model from the single searchable model picker (searchable by model name or route, with recommended models listed first). Check the selected model’s supported tools and output capabilities.
OpenRouter bills model calls to that account. Glow execution and action usage still applies. Reference-source preparation and lookup use separately billed Glow-managed embeddings.
Choosing a model
Compare models on the work your automation will perform. Model families alone do not establish answer quality, latency or cost for a particular task.
Start with the default Gemini 4 Argon, then compare a Flash model where cost and speed matter more, or an available OpenRouter model if needed. Keep the inputs, tool permissions and output requirements consistent across tests. Inspect the tool choices and answers, not just whether each run completed.
By Task
| Task | What to check |
|---|---|
| Classification, routing, tagging | Expected categories and handling of ambiguous inputs |
| Summarization, field extraction | Accuracy against the supplied text and required fields |
| Multi-step agent reasoning and tool use | Tool selection, permissions, approval points and completion |
| Long or unstructured document analysis | Source coverage, omissions and unsupported conclusions |
By Volume
High-volume automations amplify both cost and latency. An automation running 1,000 times a day makes the Flash/Pro choice roughly a thousand times over. Measure latency and recorded usage on representative inputs before choosing a model for high-volume work.
Setting it up
- Select the AI Agent step on the canvas to open the App drawer.
- Open the Brain (model) dropdown.
- Pick a model.
Different AI Agent Steps in the same automation can select different models. Evaluate each against its task rather than assuming one model is best for every branch.
API keys
Managed Gemini models do not require your own API key. OpenRouter requires a workspace connection; provider connectors require their own account setup.
If you would rather use your own provider, the App drawer carries connectors for OpenAI, Anthropic, Mistral and the rest. Those take your key and bill to your account with that provider.
- Run
- Select an AI Agent Step on the canvas and open the Brain (model) dropdown
- You should see
- The dropdown lists Gemini 4 Argon (Default), managed Flash and Pro models, and the OpenRouter provider switch.
What’s Next?
- 👉 AI Prompt →: generate answers using managed Gemini models.
- AI Agent: build autonomous, tool-using logic with custom model selection.
- Choosing an AI Step: compare AI Prompt, AI Data Transform, and AI Agent trade-offs.