Cognition's SWE-2
Product Hunt Rohan Chaubey ● Covered by 2 sources
Cognition launched SWE-2, a new coding model tuned for cost and capability. It scores well against bigger names while using less money and fewer turns.
Based on reporting by Product Hunt, Rohan Chaubey — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Cognition has released SWE-2, a coding model that was post-trained from Kimi K3 with reinforcement learning aimed at squeezing cost and capability at the same time. The pitch is blunt: do more work, spend less doing it.
On FrontierCode 1.1 Main, SWE-2 reaches 50.0%. Cognition says that puts it within a point of Fable 5.1 while costing 64% less, and within a few points of GPT-6 Astra at a quarter of the cost. That is the kind of comparison that gets attention because it’s not just about a score. It’s about what that score costs to produce.
The company also says SWE-2 beats SWE-1.7 on efficiency. It takes 58% fewer turns, costs 81% less, and still scores higher. That combination is the real story here. Coding models are being judged less like shiny demos and more like workers with a budget.
SWE-2 is available now in Devin Desktop and CLI. So this is not a lab-only flex. Cognition is pushing the model into the products where those savings, if they hold up in real use, will actually matter.
My take — AI-written commentary, not fact-checked reporting
This is the right religion for coding models: cheaper, faster, good enough, then better. The AI industry keeps pretending bigger is automatically smarter, which is a convenient story until the bill arrives. SWE-2 sounds like a model built for the adult table, not the demo stage.
Read more about this at: Product Hunt