Breaking

SpaceXAI Releases Grok 4.7 as Developers Weigh Coding Gains Against Higher Task Costs

September 22, 2026 at 11:46 EDT

Foundation Models
AI Agents
Software Dev & Coding

SpaceXAI released Grok 4.7 on September 21, 2026, pitching it as its strongest model for coding, agentic work and knowledge tasks. The launch also sharpened a practical question for developers: the token prices match Grok 4.6, but completed tasks can still cost more if the new model thinks longer and emits more tokens.

September 21, 2026 · SpaceXAI

Grok 4.7 raises the workload-cost question

The new model keeps Grok 4.6's standard API token rates, but measured task cost rises when longer reasoning and larger outputs are counted.

$2/$6

standard input/output per 1M tokens

500k

context window tokens

Intelligence Index score at high

53.1

output tokens per second at high

Task cost, drawn to scale

Weighted cost per Intelligence Index task, high effort. The taller column is Grok 4.7.

Takeaway: the token sticker price is unchanged, but the measured task bill is about 47% higher at high effort.

Fast mode is a separate trade-off

Availability

Cursor and Grok Build only

Rate

2x standard token pricing

API status

not on the public API

Why teams will test it

Better company-reported coding and terminal scores
Long-task training and self-checking focus
Works across API, Cursor and Grok Build

What teams must measure

Cost per successful completion
Latency on representative agent loops
Retries and human intervention

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.