The State of Model Routing — NVIDIA, Cognition, OpenRouter

Delegating work to a cheaper model can still waste money if every handoff repeats the same context. In this model-routing panel, Cognition describes Devin Fusion as keeping a capable model responsible for planning while an implementation model performs delegated work.

Its implementation model persists as a sidekick with an ongoing conversation, so earlier context can be reused. The main model receives compact reports and file references instead of every intermediate detail. This explains the design's intended saving: reduce repeated processing while keeping stronger reasoning available as a coding task becomes harder. The panel presents an architecture and company experience, rather than evidence that every multi-model workflow costs less.

“we're reducing the cost of Fable level intelligence by 40%. The way we do that is we allow Fable to still do like the planning and the the hard decision making”

Walden Yan

“Like if you run terminal bench on Opus and Haiku, like Opus will do about three times better at 1/10 the cost of Haiku, even though Haiku's significantly cheaper per token.”

Alex Atallah

“you look at like the top model being used by dollar spent on classification tasks, well, guess what it is. It's Opus.”

Alex Atallah

“I would like never recommend using like these models past like 200K tokens, under 100K if you can.”

Walden Yan

“you kind of just always have this like main frontier agent that's watching, even if it's not the one doing the work.”

Walden Yan