Zed needs to start winning deals for free LLMs

[MRiabov]

Hello,
This is just to say: I am a current user of windsurf, and I use freely available LLMs heavily here. Over the last month, my statistics are:

2500 messages sent over the last month, 90% of that to free models. Why? Grok code fast in particular is very fast; I can run 3-4 concurrent conversations with little failure, not much worse than longer-thinking models. While I'm typing this I have 3 specs being implemented concurrently, for 0 cost.

To me, Grok Code Fast is something that I use 95% of the time, only using GPT-5 or claude for planning tasks (by the way, see Github's spec-kit)

Zed should try to win contracts for free models like Codex and Grok Code Fast. It'll make it's adoption faster.