@cognition: Introducing SWE-2, our closest model yet to the frontier. On leading evals, it scores on par with recent frontier model…
Summary
SWE-2 is an AI model that matches frontier performance at up to 70% lower cost through scaled reinforcement learning.
View Cached Full Text
Cached at: 09/10/26, 08:23 PM
Introducing SWE-2, our closest model yet to the frontier.
On leading evals, it scores on par with recent frontier models – at up to 70% lower cost.
We scaled RL to multiple trillions of parameters, with a refined recipe that pushes the Pareto curve on both capabilities & cost. https://t.co/FjBxrikExs
Similar Articles
SWE-1.7: Frontier Intelligence at a Fraction of the Cost (22 minute read)
Cognition launches SWE-1.7, a highly capable AI model for agentic software engineering that achieves frontier-level performance at reduced cost, with improvements in RL training, multi-cluster infrastructure, data curation, and self-compaction for long tasks.
Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
Cognition launches SWE-2, an advanced coding model that achieves competitive performance with Fable 5.1 and GPT-Astra at a fraction of the cost, leveraging novel reinforcement learning to optimize the cost-performance frontier.
Cognition's SWE-2
Cognition's SWE-2 is a new coding model post-trained from Kimi K3 with reinforcement learning, achieving 50.0% on FrontierCode 1.1 Main and offering cost savings of 64% compared to Fable 5.1 while being competitive with GPT-6 Astra. It is now available in Devin Desktop and CLI.
@silasalberti: Demand for SWE-2 in Desktop & CLI is unprecedented. Glad to announce, we were able to secure additional compute. We're …
Cognition announces SWE-2, an AI model that matches frontier models in performance at up to 70% lower cost, and is making it free for Devin Cloud subscribers until October 8 due to high demand.
Cognition's SWE-2 achieves 92.8 on Terminal-Bench 2.1
Cognition's SWE-2 model, post-trained from Kimi K3, achieves a score of 92.8 on Terminal-Bench 2.1, offering competitive performance with lower cost compared to frontier models like Fable 5.1 and GPT-6 Astra.