Compare and save settings

In Experiments, choose workloads and run a baseline. Change settings and add combinations, or run Baseline + single options. Discover usable context runs first: quick load-only checks find the largest window the engine accepts. Context discovery is the only option enabled by default. Tick Also test full launch window to confirm that window with one long prompt, and select workloads to add cold speed samples after it.

Inspect completed results and their evidence before loading a configuration into Launch. Click Save my settings to keep it. Launch and experiment drafts are separate; loading a result does not save it or restart the model. Export results as CSV for comparison or JSON for the full record.

See Experiment and model settings for each workload, budget and customization control, including cold versus warm measurements and how to interpret the results.

Try in Launch copies the experiment settings into Launch. Expand a row with the ▸ control to reach it, together with Copy row and Delete run…. The Context choice defaults to Tested, which applies the largest successful context, keeping the slot count. 90% selects the smaller, rounded-down estimate instead; Original keeps the experiment’s own context. The choice appears only when a successful context measurement exists. Loading does not start the model or save preferences; review Launch, then save or restart explicitly.

Manage experiment history

Experiment history contains individual runs, with one row per sample. Expand a row and choose Delete run… to remove that entire run, including all its samples, logs and context probes. Delete failed / cancelled… selects every run whose overall status is failed or cancelled. It keeps completed runs even when an individual sample or quality check failed, and keeps interrupted runs unless you delete them individually.

Both actions show a confirmation with the number of runs and samples. Deletion is permanent; export full JSON first if you want to keep the records. Saved settings, their saved provenance, model files and the running model are kept. Deletion is unavailable while an operation is active.

To have a coding agent run the comparison for you through the panel’s API, see Let a coding agent find your best settings.