After crossing the billion dollar mark in annualized revenue, the start-up is reducing its rates on all its modes and promoting its own SWE-2 model, which is 64% less expensive to run.
Cognition has tightened the pricing of its developer Devin across its lineup: Fusion and Normal modes are down 30-40%, Ultra is down 15-20%, and Devin Review is down to 70%. The move comes just after the product surpassed $1 billion in annualized revenue, an amount the company says it doubled in four months.
Two technical building blocks underlie these declines. On September 10, the start-up launched SWE-2, its in-house model, which it places at 50.0% on the main track of its evaluation and 92.8% on Terminal-Bench 2.1, for an execution cost that it announces is 64% lower than that of Fable 5.1. The next day, Fusion arrived in the Devin CLI: a two-model setup that entrusts planning to a frontier model and execution to the cheaper SWE-2.
On the FrontierCode 1.1 benchmark, Cognition places Fusion at 68.8 points for an average cost of $0.60 per job. The assembly picks up SWE-2, Opus 5.5 and the GPT-6 family as needed. This price movement follows shortly after a $2 billion raise valuing the company at $48 billion.
Read the full article on the original source
Go to the source →