Introducing SWE-2 Flash, our new fast & efficient coding model. On...
Cognition@cognition
9 views
Nov 05, 2026
~1 min read
Advertisement
1
Introducing SWE-2 Flash, our new fast & efficient coding model.
On leading evals, it outperforms all American open-weight models, even ones 3x larger.
To balance transparency and cyber risks, we are building a program to securely share the weights with partners & researchers.
On leading evals, it outperforms all American open-weight models, even ones 3x larger.
To balance transparency and cyber risks, we are building a program to securely share the weights with partners & researchers.
2
SWE-2 Flash is based on @thinkymachines Inkling-Small, an American model with 276B total and 12B active parameters. The small size, combined with our highly optimized inference stack allows us to offer it at a 5x lower cost and 3x higher speed than SWE-2.
3
For the first time, we also added a mid-training stage before RL. We trained on several trillion tokens of code, scientific articles, blogs, and more, saving the scarcest, highest-quality data for last.
4
Our training proceeded with large-scale RL. While we made some tweaks to our RL recipe from SWE-2, most of the gains came from higher-quality, more diverse RL tasks.
Our research team carefully audited the instructions, environment setup, and verifiers for our tasks, while also curating several new task types.
Our research team carefully audited the instructions, environment setup, and verifiers for our tasks, while also curating several new task types.
5
SWE-2 Flash is free until the end of this month for all Pro, Max, and Teams users in Devin Desktop and Devin CLI.
Download: devin.ai/desktop
We will be sharing more details about the secure weight access program in the coming weeks.
Download: devin.ai/desktop
We will be sharing more details about the secure weight access program in the coming weeks.
6