the race for AGI is doomed.
We need small models for specific task. With the rise of AI agent I want my opencode to have access to a range of small models, max 30B parameters, each of them is good at one thing, one language, and one environment, ie frontend/backend, documentation, planning, repository scanning.
the rust foundation should provide a model for their language, divided by purpose, ie wasm, backend etc...
you would have larger context windows, and more speed, and you could have an instance of each model locally.
This is where AI becomes actually useful, and affordable