Powering open infrastructure for autonomous agents
cloud.google.com·3d ago
Cloud & Infrastructure
The New Stack·2d ago·5 min read
Cognition turned an open model into a coding beast. But it won’t release the weights.
TL;DR
Cognition has developed SWE-2 Flash, a coding model built on the open-source Inkling-Small, but is withholding its weights due to cybersecurity concerns. This new model is more cost-effective and faster than its predecessor.
✦ Why It Matters
Consider exploring open-source models like Inkling-Small for your own AI projects.
Key Takeaways
- Evaluate SWE-2 Flash for cost-effective coding solutions.
- Monitor Cognition's upcoming weight-sharing program for potential collaboration.
- Stay informed about cybersecurity implications of AI models.
- Test SWE-2 Flash before the free access period concludes.
How It Works
Cognition enhanced the performance of the open-source Inkling-Small model by incorporating a mid-training stage before reinforcement learning, which significantly improved its coding capabilities.
⚠ The Catch
Cognition's decision to keep SWE-2 Flash's weights private limits broader access to its improved capabilities, despite being built on an open-source foundation.