Google DeepMind Urged to Open-Source Legacy Models

This is less a launch than a pressure test: the community is signaling that older proprietary models still have value as public artifacts, even after Google has moved on to newer generations.

• –Google has already shifted to newer Imagen and Gemini releases, so the debate is really about whether older weights should be treated as archival infrastructure or kept closed
• –Open-sourcing would be most useful for researchers, hobbyists, and fine-tuning work, not for chasing SOTA performance
• –The Grok 1 comparison is rhetorical leverage: users are using xAI’s move to argue Google should match it
• –Any real release would need licenses, model cards, and safety constraints; without that, “open source” is mostly a slogan
• –My read: this is a useful signal of demand, but a weak predictor of an actual Google release