TopolearnThe model sets an agent's quality, speed and cost; picking well avoids paying for power you don't need.
Every agent runs on an AI model (a large language model from a provider such as Anthropic, OpenAI or Google). You choose it from the model dropdown on the Agent tab. Three kinds of choice:
- Auto: Gumloop's model router picks a model for each message based on how hard the task looks, which tools it needs and how easy the answer is to check. It can switch to a stronger model mid-task. The chat shows which model actually ran.
- Presets: Balanced for everyday work (this preset was called "Recommended" until October 2026) and Smartest for complex reasoning and long multi-step jobs. Gumloop updates the model behind each preset over time.
- A specific model by name, when you need predictable behaviour or a particular strength, such as reading images. The picker in the product is the current list; model names change often.
Some models offer extra controls: a reasoning or "thinking" setting (the model works through the problem step by step before answering, which is slower but better for hard tasks), and Fast speed modes.
How to choose:
- Start with Auto or Balanced.
- Move to Smartest only if the agent makes reasoning mistakes on real tasks.
- Cheaper, faster models suit simple, high-volume jobs like sorting emails.
Cost depends on the model and on how much text it reads and writes (measured in tokens). Triggers can use a different model from chat, so a scheduled background job can run on a cheaper model.
Remember
- Auto routes each message to a suitable model
- Balanced (formerly Recommended) for everyday work, Smartest for hard reasoning
- Stronger models cost more; upgrade only when real tasks need it