streaming-ui-lab

Four ways to render a streamed LLM markdown response, measured live on this page. Run a mode, type in the probe input while it streams, and compare the numbers.

Render mode

NaiveEvery chunk updates state; the full text is re-parsed with react-markdown.ThrottleChunks are buffered in a ref and flushed to state every 50 ms.MemoizeText is split into blocks; finished blocks are memoized, only the last re-renders.StreamdownThe streamdown package, built for streaming markdown.

SettingsSampleChunk size (chars)Interval (ms)Runs per mode

Type here while streaming

Keystrokes here feed the input delay and event duration metrics.

Output

Press Run to start streaming.

Results

Metrics averaged per mode and settings. n is the number of runs in each average.
Mode	Sample	Chunk	Interval (ms)	n	Commits	Render total (ms)	Render avg (ms)	TBT (ms)	Worst input delay (ms)	Worst event duration (ms)	FPS avg / min	Duration (ms)
No finished runs yet.