Loading
LoadingCheaper Inference / Optimization / Techniques
Status guide / 01
Available techniques have a shipped implementation. The workspace control surface shows what can run on your traffic and the tradeoffs attached to it.
Preview techniques are ideas under evaluation. They do not run on production traffic until an implementation is released and made available to the workspace.
Optimization registry / 02
Pick one. See the request before, the transform, and the saving.
Before
740 output tokensExplain your reasoning in detail.Restate the question before answering.
After
382 output tokensBe concise.Answer directly without preamble or repetition.
Applied the opt-in concise instruction
Appends a terse-output system instruction so the model answers directly without preamble or restating the question.
~48.7% shorter responses (CCoT); task-dependent risk
Concise Chain-of-Thought (arXiv 2401.05618), −48.7% response length