Price Grok 4.7 standard and Fast usage in Grok Build and Cursor

[starkdmi]

Summary

Add Grok 4.7 pricing for Grok Build and Cursor, including standard and Fast modes, provider-specific long-context thresholds, and model aliases. Grok Build attributes a Fast inference to Fast when prompt and turn evidence agree, even if its session summary also lists the standard model. Regression coverage uses synthetic fixtures.

The pricing ruleset advances from 2 to 3 so persisted usage is repriced. Separately, the Grok scan-cache parser revision advances from v22 to v23 so existing Grok Build sessions are rescanned with the corrected inference attribution. Cursor CSV rows remain session aggregates: their API-equivalent estimates use short-context rates as a documented floor unless request-level evidence is available.

Pricing

• xAI/Grok Build standard: $2/M input, $0.50/M cached input, $6/M output below 200k prompt tokens; 2x rates at or above 200k. Fast doubles the corresponding standard rates.
• Cursor standard: the same short-context rates, with 2x rates above 256k prompt tokens when a single request is known. Cursor Fast uses $4/M input, $1/M cached input, $12/M output at short context, and $6/M, $1.50/M, $18/M above 256k when a single request is known.

Validation

• Added pricing boundary, normalization, and Grok Build attribution regression tests using synthetic fixtures.
• ./scripts/rust-ci.sh full passed locally, including workspace tests, macOS target checks, and daemon watch checks.

Pricing sources: xAI Grok 4.7, Cursor Grok 4.7.