Breaking

SpaceXAI's Grok 4.7 Enters Agent Arena on Launch Day, at Grok 4.6 Token Prices

September 21, 2026 at 12:43 EDT

Foundation Models
AI Agents

Grok 4.7, SpaceXAI's newest model for coding and knowledge work, is now available for community evaluation in Agent Arena, and in Battle Mode for text, vision, code, and documents. The listing landed on September 21, 2026, the same day SpaceXAI published the model and priced it at the same token rates as Grok 4.6.

Sep 21, 2026 · SpaceXAI

Grok 4.7 enters live agent evaluation at Grok 4.6 prices

The new coding and knowledge-work model is open for long-horizon tool use. Its API rate is unchanged: $2 per million input tokens and $6 per million output tokens.

Input price per million tokens

Same $2 rate as Grok 4.6. Half of GPT-5.6 Sol, and one-fifth of Fable 5.1, on SpaceXAI's launch table.

Grok 4.7

Grok 4.6

GPT-5.6 Sol

$10

Fable 5.1

Each block is $1 per million input tokens. Output is $6, $6, $20, and $50 on the same table.

46.3%

CursorBench 4.0, vs 40.4% for Grok 4.6 and 51.8% for Fable 5.1

64.0%

EEBench, ahead of Grok 4.6, GPT-5.6 Sol, and Fable 5.1

500K

Token context window; knowledge cutoff May 2026

What Agent Arena measures

1 · Live task

→

2 · Search, files, terminal

→

3 · Five outcome signals

→

4 · Score vs average

Where the lab table is stronger

Up from Grok 4.6 on every published score in the launch table, including Terminal-Bench 4.0 at 38.0% versus 20.3%, at the same token price.

Where it still trails

Behind Fable 5.1 on CursorBench, terminal work, briefcase, and HealthBench. The Sep 15 Agent Arena board does not yet list Grok 4.7.

Earlier Grok models sat mid-pack before this listing

On the September 15 snapshot, Grok 4.5 was +2.92% net improvement and Grok 4.6 (xHigh) was +2.01%, versus +13.71% for the leader. Those are not Grok 4.7 scores.

Continue reading

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