Cognition AI co-founder Walden Yan launches Devin Fusion, a hybrid-model routing system that cuts coding agent costs by 35%

especially with fable-level models, i'm *very* bullish on mid-session routing.

(e.g. more routing via sub-agents, etc.)

Conventional model routing sucks. It passes benchmarks but fails to write code you'd actually merge.

Introducing Devin Fusion, a new hybrid-model harness for agentic coding.

In testing, it reduces the cost of Fable-level intelligence by 35% and still feels good to use.

For more research on how we built Devin Fusion, including examples, check out our blog:

https://cognition.com/blog/devin-fusion

Devin Fusion uses a hybrid-model harness built around two ideas:

First, a “sidekick” agent: a smaller agent runs in parallel with the frontier agent. The frontier agent delegates work, monitors progress, and keeps ownership of planning, ambiguity, and final review.

This lets Fusion stay intelligent while avoiding unnecessary frontier-model spend.

We developed FrontierCode to try to measure whether you’d actually merge a PR to production.

Devin Fusion builds on this research while avoiding common pitfalls of model routing, like expensive cache misses and lack of generalization.

Second, Fusion routes dynamically mid-session.

Engineering tasks often reveal their difficulty late: a “simple bug” might become a multi-file race condition investigation. Fusion can shift work back toward the smarter model when needed, including around context-compaction moments where switching can be much cheaper.

Devin Fusion is available today in Devin. Try it out:

http://app.devin.ai/signup?source=fusion_socials

@cognition FrontierCode is not an easy bench to win - let's go!

@cognition we are finally in the model routing era they all told me so much about

@cognition bullshit benchmarks and bullshit results, just trying to pump their own software's revenue while Fable is disabled.

@cognition @grok how is this different than a subagent or a simple orchestrator/routing architecture?

@cognition Awesome work guys! On the graph, is each point a model or a reasoning effort? I'm assuming the latter? For anyone wanting to learn more about how model routing works, we've written a guide here: https://www.notdiamond.ai/blog/a-comprehensive-guide-to-model-routing

@cognition Like it

@cognition sidekick is actually such a clever cache and context optimization over subagents

@cognition Wow huge!!

@benhylak dumb q but wouldn’t that suck in terms of kv cache invalidation costs?

@cognition I will definitely be trying this out today!!

@cognition awesome work!!

fusion has been achieved internally