Calls vs. intelligence

Sources & assumptions

OpenCode uses spending limits: 20% of a model’s monthly allowance per five-hour window and 50% per week. Its request estimates use different token and cache profiles per model. Your actual calls can be lower, especially with long outputs or heavy reasoning.

Models outside the Go list have no Go request allowance and are shown only in the intelligence ranking. Reasoning variants of matched Go models are grouped in their variant selector. Scores describe the selected Artificial Analysis benchmark variant, not a measurement of OpenCode’s serving configuration. Automatic matching uses model identity, never a similar model’s score. You can choose another benchmark in model details; manual choices are saved in this browser.