Models & Reasoning Effort
Trace isn't locked to one model. Pick whichever one fits the task from the model picker in the side panel, and switch any time — mid-conversation included.
Choosing a model
The model catalog covers a wide range of providers — including Claude, GPT, Gemini, DeepSeek, Qwen, and GLM — and stays current as new models ship, since it's fetched live rather than fixed to a hardcoded list. Different models trade off speed, cost, and reasoning depth differently; if one feels slow or a specific model briefly says it's at capacity, switching to another is usually the fastest fix (see Troubleshooting).
Managed access vs. your own API key
By default, Trace routes model requests through its own managed backend, so you don't need to set anything up to start. If you'd rather use your own API key — for your own billing, rate limits, or a provider not included by default — add it in Settings, and Trace uses that key instead.
Reasoning effort
On models that support it, reasoning effort controls how much the model "thinks" before it acts — from a quick, low-effort pass to a slower, more thorough one. Lower effort answers faster and is fine for small, well-defined changes; higher effort is worth the extra time on anything that needs real planning, like a multi-file refactor or debugging a subtle issue.
Set it from the effort selector next to the model picker. It applies to your very next message, and only has an effect on models that actually support adjustable reasoning — on models that don't, the setting is simply ignored rather than causing an error.