AI SDK 6.0.330 incrementally validates streamed arrays instead of reprocessing the full accumulated output on every chunk
Vercel released AI SDK 6.0.330 on November 11 to fix excessive CPU use while streaming long arrays through Output.array or streamObject. The parser now processes and validates completed elements incrementally rather than repeatedly walking the entire accumulated array for every text chunk.
Why it matters
Lists, tables, and data cards often stream structured array items into generative interfaces. Re-scanning all prior output on each chunk can make processing grow toward quadratic cost as the array expands, increasing server or main-thread CPU, UI stalls, and latency. Treating completed elements as the incremental state boundary better matches streamed UI.
A closer look
The fix is scoped to the AI SDK 6.x ai@6.0.330 maintenance release and the long-array streaming path for Output.array and streamObject. The release does not claim a public API or output-schema change.
Who this is for
Frontend, Node service, and platform teams using AI SDK 6, Output.array, streamObject, or generative UI with long lists and tables.
Original sources
Revisions & corrections (1)
— The immutable AI SDK maintenance release documents a specific structured-streaming CPU regression and its incremental parsing fix, directly relevant to long generated lists and data-driven interfaces.