Skip to main content
Latest Update
2026-11-04 12:02 UTC

Cognition: SWE-2 Flash Post-Trained on Inkling-Small

Cognition: SWE-2 Flash Post-Trained on Inkling-Small

Cognition announced SWE-2 Flash, its fast coding model post-trained on Inkling-small, which outperforms all American open-weight models on leading evals despite smaller size and will be shared securely with select partners.

Source

Analysis

Cognition disclosed that SWE-2 Flash is post-trained on Inkling-small, delivering higher scores than every American open-weight model on leading evals, including those three times larger, while the lab rolls out controlled weight access for partners and researchers to manage cyber risks.

Soumith Chintala

@soumithchintala

Cofounded and lead Pytorch at Meta. Also dabble in robotics at NYU.