You do get $10 worth of monthly claude credits for free - https://zed.dev/blog/zed-ai-billing but for others like OpenAI or DeepSeek you need an API key.
One thing that’s not in the article: using your own endpoints can be a bit bothersome, even if you have (like me) a LiteLLM proxy to abstract away all your keys. Zed has trouble setting that up (especially with authentication) and seems to be unable to set custom models to “reasoning” mode and hide or compact the <think> tags, which makes for a messy chat window.
Sadly, neither the distilled model nor the “full” ones seem to be able to actually work their way through a real coding problem without a lot of frustration on my part. You’d expect fetching and parsing web content to be something they “understood” down to the fundamentals…
I really like Zed, especially its attitude of starting from the scratch.
But if you want to be commercially successful, maybe it's worth providing some free trials. In comparison, Cursor has 50 trials.
At least it can give people more motivation to download.
Haven't looked into the detail but a privacy mode is also important for providing R1.
You do get $10 worth of monthly claude credits for free - https://zed.dev/blog/zed-ai-billing but for others like OpenAI or DeepSeek you need an API key.
One thing that’s not in the article: using your own endpoints can be a bit bothersome, even if you have (like me) a LiteLLM proxy to abstract away all your keys. Zed has trouble setting that up (especially with authentication) and seems to be unable to set custom models to “reasoning” mode and hide or compact the <think> tags, which makes for a messy chat window.
Can we use an api key that doesn’t involve sending our code to China with deepseek?
You can now deploy it in Azure. I’m using that now to replicate what I did with Ollama the other day: https://taoofmac.com/space/blog/2025/01/29/0900
Sadly, neither the distilled model nor the “full” ones seem to be able to actually work their way through a real coding problem without a lot of frustration on my part. You’d expect fetching and parsing web content to be something they “understood” down to the fundamentals…
Yes, you can use open router [1] and decide where to send your code.
[1] https://openrouter.ai/deepseek/
https://zed.dev/blog/how-is-deepseek-r1-for-coding#running-d... is one option