ai@7.0.136
Vercel AI SDK 7.0.136 stops chunk and first-chunk timeout timers once the model response ends, preventing long-running local tool execution from causing false model output timeouts.
Why this signal matters
In version 7.0.136 of the Vercel AI SDK (patch release 7b5497c), streaming output timeouts were adjusted to prevent long-running local tool executions from triggering output timeouts. The timers for chunkMs and firstChunkMs now stop when the model response ends, while stepMs continues to cover the entire step including local tool execution. Additionally, retries are now allocated fresh output timeout budgets. The release also updates the dependency @ai-sdk/gateway to version 4.0.110.
Actionable summary
ai@7.0.136 resolves streaming output timeout issues: chunkMs and firstChunkMs now stop upon model response completion so local tool calls do not trigger timeouts, retry attempts receive reset timeout budgets, and stepMs remains in effect across the entire step.
- Agent usefulness
- 77/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade to ai@7.0.136 if using streaming operations with long-running local tool executions that were incorrectly hitting chunkMs or firstChunkMs timeouts.