Key Info
OpenCode deliberately computes tokens per second in a conservative way so that model providers, including its own team, cannot inflate performance numbers. Because there is no industry agreement on the metric, other agent harnesses calculate it differently, leading users to report that the same model feels slower in OpenCode.
Highlights
- OpenCode uses a conservative tokens-per-second calculation to avoid provider cheating.
- There is no shared standard for how harnesses compute this metric.
- Other harnesses compute tokens per second differently, producing higher figures.
- As a result, users perceive the same model as slower when running in OpenCode.