Breaking

SpaceXAI releases Grok 4.7 for coding and knowledge work at Grok 4.6 prices

September 21, 2026 at 13:18 EDT

Foundation Models
Software Dev & Coding
AI Agents

SpaceXAI on Monday released Grok 4.7, which it calls its most capable model for coding and knowledge work, at the same standard price and speed as Grok 4.6. It is available the same day in Cursor and in Grok Build, on the xAI API as grok-4.7, and is beginning a gradual rollout in GitHub Copilot.

SEPTEMBER 21, 2026 · SPACEXAI

Grok 4.7 keeps the Grok 4.6 price

SpaceXAI's coding and knowledge model posts higher scores than Grok 4.6 on long agent tests, without taking the lead on every benchmark.

46.3%

CursorBench 4.0, up from 40.4%

71.0%

DeepSWE v1.1 at high effort, up from 65.2%

$2 / $6

Input / output per million tokens, same as Grok 4.6

3.3%

Risky dual-use prompts let through on HackerBench v0.3

Output list price, per million tokens

Each block is $2. Column height matches the published output rate.

Grok 4.7

$20

GPT-5.6 Sol

$50

Fable 5.1

Fable 5.1's output rate is a bit more than eight times Grok 4.7's. Grok 4.6 matches Grok 4.7 at $6 and is omitted.

CursorBench 4.0

Each block is about 2 percentage points. Labels are the published scores.

40.4%

Grok 4.6

41.7%

GPT-5.6 Sol

46.3%

Grok 4.7

51.8%

Fable 5.1

Grok 4.7 clears Grok 4.6 and GPT-5.6 Sol. Fable 5.1 remains ahead.

Where it gains

Same $2 and $6 list rates as Grok 4.6, with higher CursorBench, Terminal-Bench, EEBench, legal-agent, and office-work scores in SpaceXAI's table. Output is far cheaper than Sol at $20 and Fable 5.1 at $50.

Where it still trails

Fable 5.1 leads CursorBench and Terminal-Bench. GPT-5.6 Sol leads DeepSWE and HealthBench. Independent testing also found much higher output-token use, so a task can cost more than the list rate suggests.

The $2 / $6 rate is the short-prompt price

On the API, prompts of 200,000 tokens or more bill at $4, $1 cached, and $12. Fast doubles speed and token rates, and it is limited to Cursor and Grok Build. Output is text only. The 71.0% DeepSWE figure is a high-effort score, not the xhigh setting used for the other Grok 4.7 cells.

Continue reading

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