Breaking

Cognition Cuts Devin Costs and Says Devin Fusion Now Leads FrontierCode 1.1 Extended

September 28, 2026 at 14:14 EDT

AI Agents
Software Dev & Coding

Cognition said on September 28, 2026 that Devin is becoming materially cheaper while its multi-model Devin Fusion harness now leads FrontierCode 1.1 Extended. In its product update, the company said Devin is now 30-40% cheaper in Fusion and Normal mode, 15-20% cheaper in Ultra, and up to 70% cheaper in Devin Review.

Sep 28, 2026 | Cognition

Devin cuts the cost of agent work

Cognition says Devin is now cheaper across Fusion, Normal, Ultra and Devin Review, while Devin Fusion posts a 68.8 FrontierCode 1.1 Extended score at $0.60 per task.

30-40%

Cheaper in Fusion and Normal

15-20%

Cheaper in Ultra

70%

Maximum Devin Review reduction

68.8

Fusion score on Extended

Average cost per FrontierCode task

One block represents about $0.30. Taller columns cost more.

$0.60

Devin Fusion

$0.90

Opus 5.5 high

$2.62

GPT-6 Astra high

$2.68

Fable 5.1 medium

How the savings are supposed to happen

Pick the right model for each part

Lead model plans and reviews

Sidekick executes bounded work

Batch calls and reuse cache

Why users care

Lower average task costs can stretch the same usage budget across more pull requests, tests, migrations and reviews.

Limits to watch

Delegating subtle judgment can still fail, and caching gains depend on stable context, cache lifetime and model-provider constraints.

Continue reading

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