Breaking

SpaceXAI Releases Grok 4.7 With 38.0% on Terminal-Bench 4.0, Trailing Fable 5.1

September 21, 2026 at 22:50 EDT

Foundation Models
AI Agents
Software Dev & Coding

SpaceXAI released Grok 4.7 on September 21, positioning it as its strongest model for coding and knowledge work. The launch highlights a sharp Terminal-Bench 4.0 gain over Grok 4.6, from 20.3% to 38.0%, but the same comparison keeps Fable 5.1 far ahead at 57.9%.

Sep 21, 2026 · SpaceXAI

Grok 4.7 narrows the terminal-agent gap

The new model nearly doubles Grok 4.6 on Terminal-Bench 4.0 in SpaceXAI's comparison, but Fable 5.1 remains the stronger result on that same scale.

Grok 4.7

38.0%

Terminal-Bench 4.0

Grok 4.6

20.3%

Previous model

Fable 5.1

57.9%

Comparison leader

API price

$2/$6

Input/output per 1M tokens

Terminal-Bench 4.0, drawn to scale

20.3%

Grok 4.6

38.0%

Grok 4.7

57.9%

Fable 5.1

Each stack uses the same percent unit; taller columns mean higher task success.

Harness choice changes the reading

Grok Build comparison

38.0%

mini-swe-agent pass@1

25.8%

The practical takeaway is to compare the model, effort setting, and harness together, then validate on internal tasks.

Why teams will test it

Same starting token price as Grok 4.6, stronger long-horizon training, broad coding-tool availability.

What teams will watch

Terminal-agent gaps, output-token burn, retries, latency, and human repair on real workloads.

Continue reading

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