grok-4.7

Vendor xai Tier - Price per million tokens, in / out $2 / $6 Released 2026-09-21 Context 500,000 Max output 500,000 Vendor page

Last updated 2026-10-26 How this is made

Meant to excel at

xAI-reported strengths in electrical engineering (EEBench 64.0 vs 56.4 for Fable 5.1) and Harvey legal agent tasks (19.6 vs 6.7); vendor numbers only.
Vendor's own site x.ai
Cursor positions it for harder, longer-running multi-step agent work on large codebases, including tasks Grok 4.6 struggled with.
Vendor claim, relayed cursor.com

Watch out for

Very verbose and slow to start: about 240M output tokens to run the AA index (median 85M), 46.6s time to first token, and $3.74 per index task (rank 102 of 114).
Independent evaluator artificialanalysis.ai
Trails Fable 5.1 on terminal work (Terminal-Bench 4.0 37.6 vs 57.9, xAI-reported); Vals measured only 28.8 on the same benchmark.
Independent evaluator vals.ai
Mid-pack on blind WebDev voting (1637, rank 18) and AA index 46, well below Anthropic, OpenAI, Google and Meta flagships.
Independent evaluator arena.ai

Benchmarks

Coding percentile 0.34

Benchmark	Score	Percentile	Source	Date
code arena webdev	1637	0.29	Independent evaluator arena.ai	2026-10-19
lmarena code arena webdev	1637	0.39	Independent evaluator arena.ai	2026-10-19

Agentic coding percentile 0.26

Benchmark	Score	Percentile	Source	Date
cursorbench 4.0	46.3	0.00	Vendor's own site x.ai	2026-09-21
deepswe v1.1	71	0.40	Vendor's own site x.ai	2026-09-21
terminal-bench 4.0	37.6	0.38	Vendor's own site x.ai	2026-09-21

Overall percentile 0.40

Benchmark	Score	Percentile	Source	Date
artificial analysis intelligence index	46	0.46	Independent evaluator artificialanalysis.ai	2026-10-20
vals index	58.56	0.33	Independent evaluator vals.ai	2026-10-20

Percentile is the share of tracked models this model beats on that benchmark, averaged across the group. Only benchmarks at least four tracked models share are scored.

Practitioner reports

+/-

Cursor advises using Grok 4.7 for hard, long-running agent work and Composer for everyday coding where speed and cost matter.

undated
Vendor claim, relayed cursor.com