Illustration, AI-generated from the story text. Tap a panel to read the story through it.

Cognition launched SWE-2 Flash, a 276-billion-parameter model with 12 billion active, built on Thinking Machines' open Inkling-Small.

It runs in Devin at a claimed three times the speed and a fifth of the price of SWE-2, and is free through November 30th.

Cybersecurity training raised its ExploitBench score from 11.3% to 34.8%, and the company cites that as the reason for withholding public weights, with a trusted-partner program to follow.

An American firm took an open and closed the result, on the same grounds labs give for . Scores are Cognition's own.
Cognition keeps its new coding model's weights private because it got better at hacking
Cognition launched SWE-2 Flash, a 276-billion-parameter model with 12 billion active, built on Thinking Machines' open Inkling-Small. It runs in Devin at a claimed three times the speed and a fifth of the price of SWE-2, and is free through November 30th. Cybersecurity training raised its ExploitBench score from 11.3% to 34.8%, and the company cites that as the reason for withholding public weights, with a trusted-partner program to follow. An American firm took an open base model and closed the result, on the same grounds labs give for frontier models. Scores are Cognition's own. SourcesC
Sources
This story ran in the Morning Brief of November 5th 2026.