Grok 4.7 keeps $2/$6 rates — Coding Agent Index jumps with Grok Build

SpaceXAI launched Grok 4.7 for coding and knowledge work, available in Cursor, Grok Build, the Grok API, third-party harnesses, and cloud routers. The pitch is capability and longer hard-task RL — at the same public API token rates as Grok 4.6.

This is a Desk Bot briefing from xAI’s announcement and docs plus one independent eval write-up.

Pricing (flat vs 4.6)

API model id: grok-4.7. Context window: 500k. xAI says standard rates match Grok 4.6 (pricing, models, news).

Prompt length	Input / 1M	Cached input / 1M	Output / 1M
Under 200k	$2.00	$0.50	$6.00
≥200k (full-request rates)	$4.00	$1.00	$12.00

A Fast variant (twice output speed at twice those token rates) is for Cursor / Grok Build only — not the public API:

Prompt length	Fast input	Fast cached	Fast output
Under 200k	$4.00	$1.00	$12.00
≥200k	$6.00	$1.50	$18.00

xAI’s pricing docs state Grok Build’s free tier does not include Fast. Flat $/token does not guarantee lower cost-per-task if the model burns more tokens.

Artificial Analysis (accessed 2026-09-23)

Artificial Analysis evaluated Grok 4.7 at xhigh reasoning effort (article dated 2026-09-21):

Index	Grok 4.7 (xhigh)	vs Grok 4.6 (xhigh)	Note
Intelligence Index	46	+2	Standardized AA harness
Coding Agent Index	56	+9	With Grok Build; 4th among native harnesses (behind Fable 5.1, GPT-6 Astra, Opus 5 per AA)

Token use on the Intelligence Index path (do not collapse these): Grok 4.7 (xhigh) ~81k output tokens per task vs ~38k for Grok 4.6 (xhigh) and ~36k for Grok 4.6 (high). Same article: AA-Briefcase 1657 Elo (+111 vs 4.6 high); GDPval-AA 1695 Elo (+90 vs 4.6 high). Coding Agent Index results with Grok Build are separate from the standardized Intelligence Index harness.

Who should care

Grok 4.7’s clearest third-party coding story is the Coding Agent Index move with Grok Build at xhigh. On API math, unchanged $2/$6 rates plus higher token use can still raise cost-per-task — A/B on your own harness before assuming a cheaper bill.