Breaking

SpaceXAI Releases Grok 4.7 With Coding Gains at Grok 4.6 Pricing

September 21, 2026 at 20:01 EDT

Foundation Models
AI Agents
Software Dev & Coding

SpaceXAI released Grok 4.7 on September 21, presenting it as its strongest model yet for coding and knowledge work while keeping the same starting API price as Grok 4.6. The launch gives developers a larger-base Grok model trained for longer-running agent tasks, but early measurements also show higher token use that could matter for real workloads.

Sep 21, 2026 · SpaceXAI

Grok 4.7 Targets Agent Work

A larger-base Grok model raises coding and knowledge-work scores at Grok 4.6 starting prices. The open question is whether stronger long-task performance offsets heavier token use.

per 1M input tokens

per 1M output tokens

500k

token context window

Coding Agent Index

Token Use Became the Watch Item

Output tokens per benchmark task, shown with one block representing about 5k tokens.

38k

Grok 4.6

81k

Grok 4.7

The newer model scored higher, but the measured output-token load more than doubled in this comparison.

What Changed Under the Hood

Larger base model->Longer RL run->More self-checking->Longer agent tasks

Upgrade Case

Better long-horizon coding and office-work scores
Same standard per-token price as Grok 4.6
Available in common coding workflows

Caution Case

Higher output-token use can raise task cost
Some comparisons use different reasoning levels
Fast and 500k tiers cost more

Continue reading

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