Thinking Modes and Effort

Modes depend on the lane

Thinking modes are lane-specific. The catalog lists low, medium, and high for Borg Fusion. Borg Fusion Coding, Borg Fusion Design, and Borg Fusion Fast also list max. Do not assume that a mode listed for one profile is available on another lane. Check the model's page and its thinking modes information before sending a setting.

A mode is an effort choice, not a speed guarantee. More effort can be useful when a task needs careful planning, but a short, simple request may not need it. Start with the lowest mode that gives an acceptable result, then compare on representative tasks.

Use measured timing carefully

Borg telemetry notes report time to the first streamed response event and end-to-end completion tokens per second. These are different measures. Time to first event is not total response time. Completion tokens per second includes the whole turn, including queueing and reasoning, so it is not the rate at which each streamed token appears.

Measurements are observations from production requests over a 30-day window, not promises about the next request. The telemetry only publishes some fields when sample counts meet its threshold. A missing value is not evidence that a model is faster or slower.

Use the notes to compare likely tradeoffs, not to rank every task. For example, a model can take longer before its first event yet produce its full answer at a different overall pace. Test the same prompt and output needs on your own workload. Compare answer quality, time to first event, full completion time, and the amount of reasoning needed. For latency-sensitive tasks, begin with lower effort or a speed-focused profile. For difficult work, test higher effort and check whether the answer improves enough to justify the added wait.