Breaking

Cognition Adds Claude Fable 5.1 to Devin, Cutting Fable Task Cost 54%

September 1, 2026 at 14:07 EDT

Foundation Models
Software Dev & Coding
AI Agents

Cognition has added Anthropic’s Claude Fable 5.1 to Devin across Desktop, CLI, and Cloud, saying Fable-level intelligence is now 54% cheaper on end-to-end tasks than Claude Fable 5 and undercuts Claude Opus 5 on measured cost per completed task after a fourfold cut in cached-token prices.

September 1, 2026 · Cognition

Fable 5.1 in Devin: same intelligence, 54% less per task

Anthropic cut cache-read prices fourfold. Cognition says Devin Fusion now matches Fable 5.1 on FrontierCode at 47% lower cost.

54%

cheaper than Fable 5 per FrontierCode task

47%

Fusion discount vs Fable 5.1, matching score

cheaper cache reads ($0.25 vs $1.00 / MTok)

95%+

of coding-task tokens are cache hits

Cost per FrontierCode task

Medium thinking. Column height is proportional to dollars spent.

$5.84

Fable 5

$3.51

Opus 5

$2.68

Fable 5.1

$1.43

Fusion

Fable 5.1 undercuts Opus 5 on task cost despite a $50 vs $25 output sticker. Fusion matches Fable 5.1’s 63.2–63.6 score band at $1.43.

How Fusion spends frontier tokens

1. Plan

Frontier model owns the brief and judgment calls

→

2. Execute

Cheaper sidekick implements, tests, and iterates

→

3. Review

Frontier model checks the diff and commits

What improved

Fable 5.1 is cheaper than Opus 5 on completed tasks. Fusion holds a 63.2 score at $1.43. Users with Fable on should see 10–25% savings on real work.

What to watch

The FrontierCode score peaks at medium effort; higher reasoning adds out-of-scope diffs. Zero-data-retention access is a time-bound exemption, and Fable can still exhaust plan limits quickly.

Continue reading

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