Breaking

Cognition Adds Claude Opus 5.5 to Devin as Model Tops FrontierCode 1.1

September 22, 2026 at 13:36 EDT

Foundation Models
AI Agents
Software Dev & Coding

Cognition has added Claude Opus 5.5 to Devin, putting Anthropic's new Opus-tier model directly into a production coding-agent workflow on launch day. On Cognition's FrontierCode 1.1 Main leaderboard, Opus 5.5 now ranks first with a 54.6% score, narrowly ahead of Claude Fable 5's 53.5%, while the measured cost per task is $0.80 versus $13.09 for the Fable 5 run.

Sep 22, 2026 · Agentic coding

Opus 5.5 enters Devin at the top

Claude Opus 5.5 moves into Devin with the leading FrontierCode 1.1 Main score and a much lower measured task cost than the prior Fable 5 lead.

54.6%

FrontierCode 1.1 Main

$0.80

Measured cost per task

40%

Typical run-cost drop vs Opus 5

Token context window

Same benchmark, different cost curve

Opus 5.5$0.80

Fable 5$13.09

Each stack compares measured cost per FrontierCode 1.1 Main task; the taller column is about 16 times the spend.

Why developers care

The update pairs a first-place mergeability score with lower token prices, lower cache-read cost and reports of fewer steps on long coding jobs.

What still needs testing

The leaderboard gap is narrow, task costs depend on the harness, and production safeguards can affect restricted domains.

The practical test

For teams using coding agents, the question is whether Opus 5.5 keeps its efficiency advantage on their own repositories, review standards and long-running workflows.

Continue reading

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