Devin News 07/13/2026 AI Rating: Low

Cognition Says Pairing Fable with Its "Fusion" Sidekick Setup Beats Opus on Cost and Score

#Devin#Cognition#Fable#Opus#FrontierCode

Cognition, maker of the autonomous AI software engineer Devin, has published results showing that pairing its multi-agent “Fusion” configuration with Anthropic’s Fable 5 costs less and scores higher than running Opus 4.8 alone. The analysis is based on 3,000 evaluation sessions run on FrontierCode 1.1.

Details

  • Score and cost comparison: “Fable + Fusion” (a sidekick setup) costs $1.86 per run with a score of 60.7, versus “Opus + Fusion” at $2.04 per run with a score of 54.6 — Fable comes out both cheaper and higher-scoring
  • Versus plain Fable: Using Fable in the sidekick configuration cuts cost by 54% versus running Fable alone, while maintaining performance
  • Behavioral difference: Cognition describes it as “Opus behaves like a micromanager with an intern; Fable is a manager with a capable engineer” — Fable delegates early with clear constraints, while Opus tends to delegate only after expensive exploratory work
  • Scale of testing: The analysis draws on 3,000 evaluation sessions run on the FrontierCode 1.1 benchmark

How to try it

  • This post is a technical analysis of model-routing strategy inside Devin, not a standalone new feature or product
  • Full methodology and data are published on Cognition’s official blog