Breaking

Cognition Adds Grok 4.7 to Devin for Harder Coding-Agent Workloads

September 21, 2026 at 17:28 EDT

Software Dev & Coding
AI Agents
Foundation Models

Cognition AI has made Grok 4.7 available in Devin, saying the model scored 59.4% on FrontierCode 1.1 Extended and performed especially well on hard backend engineering tasks. The rollout places SpaceXAI's newest coding and knowledge-work model inside one of the more visible autonomous software-engineering agents on the same day Grok 4.7 was introduced publicly.

Sep 21, 2026 · Devin

Grok 4.7 moves into Devin

Cognition added SpaceXAI's newest coding and knowledge-work model to Devin, citing a 59.4% score on FrontierCode 1.1 Extended and strong performance on hard backend engineering tasks.

59.4%

FrontierCode 1.1 Extended in Devin

500k

documented context window

$2/$6

base input and output per 1M tokens

46.3%

CursorBench 4.0 score

Benchmark scale

The Devin result sits at 59.4 points on a 100-point FrontierCode Extended scale.

59.4%Grok 4.7 in Devin

100%full scale

What changes in practice

Select model

Use Grok 4.7 for a Devin coding session.

→

Run long task

Lean on larger context and higher reasoning settings.

→

Review PR

Judge the result by mergeability and testing.

Positive read

Developers are testing it for live apps, long tasks, architecture discussion, documents, and backend work.

Open concern

Higher token use and reasoning-level differences make real project cost an important test.

Continue reading

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