01 / DECIDE
Architecture decision
Use it when one product surface mixes interactive edits with analysis that approaches the 500k context window, and the resolved model identifier can change between releases.
Grok 4.7 and Grok 4.6 add an xhigh effort level above the default high, while Grok 4.5 treats an xhigh request as high, so an effort level only means something next to a pinned generation. Map input, output, state, tools, and side effects as one observable path, then name the owner of pin the model generation and the reasoning effort together.
02 / OPERATE
Three-step field procedure
- 01
Lock the contract
Write down the caller, data classification, success condition, timeout, cancellation, and ownership. Use it when one product surface mixes interactive edits with analysis that approaches the 500k context window, and the resolved model identifier can change between releases.
- 02
Run one bounded path
Build one end-to-end path with correlation, typed state, and reversible failure handling. Grok 4.7 and Grok 4.6 add an xhigh effort level above the default high, while Grok 4.5 treats an xhigh request as high, so an effort level only means something next to a pinned generation.
- 03
Verify the signal
Turn acceptance into a repeatable fixture, contract test, or browser test. Every run records the resolved model identifier, the effort level, the prompt size against the 200k pricing threshold, and the task outcome.
03 / BOUND
Operational boundary
Reasoning cannot be disabled and a higher effort level is not proof of a correct answer; the fast variant is served only through Grok Build and Cursor at twice the standard token rates, not on the public xAI API.
04 / PROVE
Acceptance evidence
Every run records the resolved model identifier, the effort level, the prompt size against the 200k pricing threshold, and the task outcome.
SOURCE / HTTP
Reproducible source probe
curl -fsSI 'https://docs.x.ai/developers/grok-4-7' | sed -n '1,5p'