Comparison
StreamingvsStreaming SSR
Streaming
the twenty-second wait becomes an answer that visibly starts in under a second, without the model getting any faster.
Receiving the response token by token as it is generated instead of waiting for the whole thing. Essential for chat UX: it converts a 20 second wait into a response that visibly starts in under a second.
Full entry →Streaming SSR
the header and layout arrive immediately and the slow part of the page fills in a moment later, in the same response.
Sending server-rendered HTML in chunks as it becomes ready instead of waiting for the whole page. The shell reaches the browser before the slowest data fetch finishes, so time to first byte stops being hostage to the slowest query. The costs are that you cannot set headers or a status code once the stream has started, and that error handling after the first byte means rendering a fallback in place rather than redirecting.
Full entry →