Cognition Launches SWE-2 Flash, a Faster and Cheaper Coding Agent

At a glance

Cognition has released SWE-2 Flash, a coding model. According to the company, it is three times faster and one-fifth the price of its larger sibling, SWE-2. It targets everyday software work such as automation, code review, and feature development, and it is free to use until November 30, 2026.

Key points

1. 1

   It does not lead in every area. It was slightly ahead on FrontierCode 2.0, but SWE-2 scored higher on Terminal-Bench 4.0 and DeepSWE 1.1.

2. 2

   The weights (the learned numerical values of the AI model) have not been released yet. The company plans to share them only with trusted partners.

3. 3

   Its cybersecurity capabilities were strengthened, and its ExploitBench score rose from 11.3% to 34.8%. Because of concerns it could be misused, the company decided not to release the weights.

Why it matters

This case shows how cheaper, faster AI coding tools may be used in real development work, and how decisions are made about the scope of releasing powerful AI.