…s v6 (#20841)

Automated update of the gateway model settings files.

Regenerated from the upstream gateway model list by
[`update-model-settings.yml`](https://github.com/vercel/ai/blob/main/.github/workflows/update-model-settings.yml)
([this run](https://github.com/vercel/ai/actions/runs/35121093565)).
Merging this PR is what surfaces newly-released gateway models as IDE
autocomplete suggestions when users type a model string with
`@ai-sdk/gateway`.

See the workflow file for details on cadence and scope.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

This PR was opened by the [Changesets
release](https://github.com/changesets/action) GitHub action. When
you're ready to do a release, you can merge this and the packages will
be published to npm automatically. If you're not ready to do a release
yet, that's fine, whenever you add more changesets to release-v6.0, this
PR will be updated.


# Releases
## ai@6.0.285

### Patch Changes

- Updated dependencies [88d0bad]
  - @ai-sdk/gateway@3.0.196
## @ai-sdk/angular@2.0.286

### Patch Changes

- ai@6.0.285
## @ai-sdk/gateway@3.0.196

### Patch Changes

- 88d0bad: Backport: chore(provider/gateway): update gateway model
settings files
## @ai-sdk/langchain@2.0.293

### Patch Changes

- ai@6.0.285
## @ai-sdk/llamaindex@2.0.285

### Patch Changes

- ai@6.0.285
## @ai-sdk/react@3.0.288

### Patch Changes

- ai@6.0.285
## @ai-sdk/rsc@2.0.285

### Patch Changes

- ai@6.0.285
## @ai-sdk/svelte@4.0.285

### Patch Changes

- ai@6.0.285
## @ai-sdk/vue@3.0.285

### Patch Changes

- ai@6.0.285

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

…uests (#20903)

This is an automated backport of #20871 to the release-v6.0 branch. FYI
@shaper
This backport has conflicts that need to be resolved manually.

### `git cherry-pick` output

```
Auto-merging content/providers/01-ai-sdk-providers/32-alibaba.mdx
Auto-merging packages/alibaba/README.md
Auto-merging packages/alibaba/src/alibaba-chat-language-model.test.ts
Auto-merging packages/alibaba/src/alibaba-chat-language-model.ts
CONFLICT (content): Merge conflict in packages/alibaba/src/alibaba-chat-language-model.ts
Auto-merging packages/alibaba/src/alibaba-chat-options.ts
Auto-merging packages/alibaba/src/alibaba-chat-prompt.ts
Auto-merging packages/alibaba/src/convert-to-alibaba-chat-messages.test.ts
CONFLICT (content): Merge conflict in packages/alibaba/src/convert-to-alibaba-chat-messages.test.ts
Auto-merging packages/alibaba/src/convert-to-alibaba-chat-messages.ts
CONFLICT (content): Merge conflict in packages/alibaba/src/convert-to-alibaba-chat-messages.ts
error: could not apply 7f803af... feat(alibaba): support preserved thinking in multi-turn requests (#20871)
hint: After resolving the conflicts, mark them with
hint: "git add/rm <pathspec>", then run
hint: "git cherry-pick --continue".
hint: You can instead skip this commit with "git cherry-pick --skip".
hint: To abort and get back to the state before "git cherry-pick",
hint: run "git cherry-pick --abort".
hint: Disable this message with "git config set advice.mergeConflict false"
```

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

This PR was opened by the [Changesets
release](https://github.com/changesets/action) GitHub action. When
you're ready to do a release, you can merge this and the packages will
be published to npm automatically. If you're not ready to do a release
yet, that's fine, whenever you add more changesets to release-v6.0, this
PR will be updated.


# Releases
## @ai-sdk/alibaba@1.0.55

### Patch Changes

- def3cdf: feat(alibaba): preserve reasoning in multi-turn requests by
default on supported models

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

…l results (#21008)

## Background

OpenAI Responses omitted provider-referenced PDF and image files from
tool results, preventing models from receiving files returned by tools.

## Root Cause

The tool-result converter handled inline data and URLs but had no
provider-reference branch, and its function output type lacked file_id
variants; the original replay confirmed references became empty output
arrays with unsupported-content warnings.

## Summary

Resolved tool-result provider references into input_file or input_image
file_id content, updated internal request types, removed reproduction
artifacts, and added a patch changeset.

## Testing

Added regression coverage for mixed text, PDF references, image
references, image detail, prompt cache breakpoints, and warning-free
conversion.

## End-to-end Validation

- `replay_original_reproduction`: completed with exit code 0 and no
original bug signal.
- `pnpm -C examples/ai-functions exec tsx -e '<live uploaded-image
tool-result validation>'`: OpenAI returned `Cat` with no warnings, and
the uploaded file was deleted successfully.

## Related Issues

Fixes #20971

Closes #21002

Backport of #20978

---------

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>
Co-authored-by: Gregor Martynus <39992+gr2m@users.noreply.github.com>

…sults and response history (#21007)

## Background

Simulated text streaming lost per-part provider metadata from final
content, response history, and continuation prompts.

## Root Cause

The text branch of simulateStreamingMiddleware created text stream
events without the generated part's providerMetadata, so stream
accumulation could not retain it; the immutable reproduction confirmed
the metadata disappeared while reasoning metadata survived.

## Summary

Forwarded generated text providerMetadata on simulated text-start events
and added a patch changeset for ai.

## Testing

Added regression coverage verifying provider metadata remains in
streamed content and is mapped to providerOptions in response messages.

## End-to-end Validation

- Immutable `pnpm -C examples/ai-functions exec tsx
src/reproduction/simulate-streaming-text-provider-metadata.ts` replay
exited successfully and retained metadata in content, response history,
and the continuation prompt.

## Related Issues

Fixes #20970

Closes #20999

Backport of #20977

---------

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>
Co-authored-by: brennanbutler01 <64561607+brennanbutler01@users.noreply.github.com>

… or time out (#21006)

## Background

Canceled or timed-out AI SDK calls could remain pending indefinitely
while downloading unsupported prompt attachment URLs.

## Root Cause

Prompt conversion did not receive the call's effective abort signal, and
the default download function therefore invoked the lower-level
downloader without a signal. The original reproduction confirmed that
all 24 downloads lacked signals before the change.

## Summary

Threaded effective abort signals through every prompt-conversion call
site and into SDK-owned attachment downloads, without changing custom
download behavior. Added an ai package patch changeset and removed
reproduction-only artifacts.

## Testing

Added regression tests covering caller cancellation for generateText and
streamText, total timeout cancellation, signal forwarding, and
prevention of model invocation after cancellation.

## End-to-end Validation

- `pnpm -C examples/ai-functions exec tsx
src/reproduction/issue-20965.ts` via immutable replay: exited
successfully; 0/24 cancellation cases remained pending, every download
received a signal, and no model was invoked before cancellation.

## Related Issues

Fixes #20965

Closes #20998

Backport of #20968

---------

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>

…al output streams (#21013)

## Background

Structured JSON streaming omitted valid null and empty-string values
from partialOutputStream, although final output resolved correctly.

## Root Cause

The stream getter treated null as missing by using a nullish check,
while duplicate suppression initialized its comparison state to an empty
string, causing the first empty-string output to appear already
published. The reproduction confirmed both paths affected streamText and
ToolLoopAgent.stream.

## Summary

Distinguished undefined from valid null output and tracked whether a
partial value was previously published independently of its serialized
value. Added an ai package patch changeset.

## Testing

Added regression coverage verifying that Output.json() streams null and
empty-string values while preserving the correct final output.

## End-to-end Validation

- `replay_original_reproduction`: the immutable streamText and
ToolLoopAgent.stream reproduction completed successfully without the
original bug signal.

## Related Issues

Fixes #20962

Closes #21010

Backport of #20973

---------

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>

…eps emit text (#21018)

## Background

Structured outputs could resolve successfully after a tool loop while
partialOutputStream and array elementStream emitted nothing if an
earlier step contained text.

## Root Cause

createOutputTransformStream retained its selected text ID, accumulated
text, metadata, and last published value across step boundaries. The
reproduction confirmed that earlier-step prose caused final-step JSON to
be skipped or parsed with stale state; resetting this state at each
start-step restored every failing case.

## Summary

Reset structured-output parsing state at each tool-loop step boundary,
retained retry isolation, removed reproduction-only artifacts, and added
an ai patch changeset.

## Testing

Added object and array regression tests covering earlier-step text,
final structured partials, completed array elements, and resolved
output.

## End-to-end Validation

- `replay_original_reproduction` (`pnpm -C examples/ai-functions exec
tsx src/reproduction/issue-20961-partial-output-across-tool-steps.ts`)
exited 0; all streamText and ToolLoopAgent object/array cases emitted
the expected partials and elements.

## Related Issues

Fixes #20961

Closes #21012

Backport of #20972

---------

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>
Co-authored-by: Gregor Martynus <39992+gr2m@users.noreply.github.com>

…th pending approval responses (#21023)

## Background

Published `ai` package behavior produced invalid pruned histories:
retained pending approval decisions lost their originating tool call,
causing text and agent APIs to throw before tool or model execution.

## Root Cause

`pruneMessages` tracked retained approval IDs separately from retained
tool-call IDs. A trailing approval response preserved its request by
approval ID but never traced the request's `toolCallId`, so the
originating call was pruned; the reproduction and failing regression
test confirmed this path.

## Summary

Added approval-to-tool-call tracing so retained approvals also retain
their originating calls, added regression coverage for general and
selective pruning, and added an `ai` patch changeset.

## Testing

Added a regression test covering pending approval histories under
`before-last-message` and selective tool pruning.

## End-to-end Validation

- `replay_original_reproduction` (`pnpm -C examples/ai-functions exec
tsx src/reproduction/issue-20953-prune-pending-approval.ts`) — exited
successfully and all 36 approval scenarios passed.

## Related Issues

Fixes #20953

Closes #21017

Backport of #20958

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>
Co-authored-by: brennanbutler01 <64561607+brennanbutler01@users.noreply.github.com>

…nt or late stale-token 401 responses (#21028)

## Background

Legacy SSE message requests redundantly refreshed OAuth tokens when
concurrent or delayed requests returned 401 responses for the same stale
bearer token.

## Root Cause

SseMCPTransport called auth() independently for every eligible 401 and
did not check whether the rejected request used an outdated token;
deterministic concurrent and after-save scenarios therefore produced two
refresh exchanges.

## Summary

Added transport-level OAuth recovery coordination and
retry-with-current-token handling, removed reproduction-only artifacts,
and added a patch changeset for @ai-sdk/mcp.

## Testing

Added legacy SSE regression tests covering simultaneous 401 responses
and a delayed stale 401 received after refreshed tokens were saved.

## End-to-end Validation

- `pnpm -C examples/ai-functions exec tsx
src/reproduction/mcp-sse-concurrent-oauth-refresh.ts` via immutable
replay completed successfully with one refresh in both race scenarios
and no bug signal.

## Related Issues

Fixes #20940

Closes #21022

Backport of #20943

---------

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>

… results and callbacks (#21026)

## Background

streamText called the model selected by prepareStep but mislabeled
completed steps and callbacks with the original model.

## Root Cause

The streaming event processor constructed DefaultStepResult and warning
metadata from the outer model even though each call used the resolved
stepModel; the reproduction confirmed model B was called while steps,
onStepEnd, and onEnd reported model A.

## Summary

Tracked the resolved model for each streaming step and used it when
constructing completed-step metadata and logging warnings. Added a patch
changeset and removed reproduction-only artifacts.

## Testing

Expanded streamText regression coverage to verify prepareStep model
changes across step-start events, step-end callbacks, returned steps,
and onEnd data.

## End-to-end Validation

- `replay_original_reproduction` (`pnpm -C examples/ai-functions exec
tsx src/reproduction/issue-20944-stream-text-prepare-step-model.ts`)
completed successfully with every streamText and ToolLoopAgent case
reporting model B and no mismatches.

## Related Issues

Fixes #20944

Closes #21016

Backport of #20946

---------

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>
Co-authored-by: Gregor Martynus <39992+gr2m@users.noreply.github.com>

…ng stops early (#21036)

## Background

Breaking iteration or cancelling a reader from the published
readUIMessageStream API could produce an unhandled closed-controller
rejection and terminate Node.js processes.

## Root Cause

Output cancellation closed the output controller while an unobserved
background consumer continued processing, allowing later writes or final
closure to target the closed controller. The original reproduction
confirmed ERR_INVALID_STATE for early exits, and the immutable replay
now confirms the failure is gone.

## Summary

The PR tracks output cancellation synchronously, aborts background input
consumption, suppresses writes and final closure after cancellation,
adds two regression tests, and includes an ai patch changeset. No
dependencies or configuration were changed.

## Testing

Node and Edge regression coverage verifies cancellation propagation for
iterator break and explicit reader cancellation. The focused tests
passed repeatedly, and the full repository tests, full type check,
formatting, and lint checks passed.

## End-to-end Validation

- `replay_original_reproduction`: exited 0 with no unhandled rejection
or original failure signal; classified as `no-longer-reproduces`.

## Related Issues

Fixes #20991

Closes #21003

Backport of #20994

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>
Co-authored-by: brennanbutler01 <64561607+brennanbutler01@users.noreply.github.com>
Co-authored-by: Gregor Martynus <39992+gr2m@users.noreply.github.com>

… type is omitted (#21041)

## Background

Amazon Bedrock responses without a body-level error type were exposed to
consumers as "undefined: <message>" in streaming chat, embedding, image,
and reranking calls.

## Root Cause

Four handlers unconditionally interpolated the nullish
AmazonBedrockErrorSchema.type field; the immutable reproduction
confirmed all four produced the literal "undefined:" prefix.

## Summary

Added a shared Bedrock failed-response handler that conditionally
includes the error type, adopted it across the affected handlers,
removed reproduction artifacts, and added a patch changeset.

## Testing

Added regression tests covering error bodies both with and without a
type, preserving typed prefixes while returning unprefixed messages when
type is absent.

## End-to-end Validation

- `replay_original_reproduction`: original streaming chat, embedding,
image, and reranking reproduction exited successfully without the bug
signal.
- `pnpm -C examples/ai-functions exec tsx
src/reproduction/amazon-bedrock-error-message-live.ts`: live AWS Bedrock
request returned `The provided model identifier is invalid.` without an
`undefined:` prefix.

## Related Issues

Fixes #20926

Closes #21038

Backport of #20933

---------

Co-authored-by: ai-sdk-factory[bot] <305873210+ai-sdk-factory[bot]@users.noreply.github.com>
Co-authored-by: ai-sdk-factory <308175966+ai-sdk-factory@users.noreply.github.com>

This PR was opened by the [Changesets
release](https://github.com/changesets/action) GitHub action. When
you're ready to do a release, you can merge this and the packages will
be published to npm automatically. If you're not ready to do a release
yet, that's fine, whenever you add more changesets to release-v6.0, this
PR will be updated.


# Releases
## ai@6.0.286

### Patch Changes

- 8a9722e: fix(ai): preserve provider metadata when simulating text
streams
- 6a0f2d9: fix(ai): report the prepareStep model in streamed step
results
- db02971: fix(ai): prevent unhandled rejections when UI message stream
reading stops early
- 4d91999: fix(ai): preserve tool calls required by retained pending
approvals
- b2ad994: fix(ai): cancel prompt attachment downloads when model calls
are aborted or time out
- 9c85bd4: fix(ai): stream null and empty string JSON partial outputs
- becaa32: fix(ai): stream structured output from the final tool-loop
step
## @ai-sdk/amazon-bedrock@4.0.178

### Patch Changes

- 96b6d3f: fix(amazon-bedrock): preserve API error messages when the
error type is omitted
- Updated dependencies [e53d917]
  - @ai-sdk/openai@3.0.114
## @ai-sdk/angular@2.0.287

### Patch Changes

- Updated dependencies [8a9722e]
- Updated dependencies [6a0f2d9]
- Updated dependencies [db02971]
- Updated dependencies [4d91999]
- Updated dependencies [b2ad994]
- Updated dependencies [9c85bd4]
- Updated dependencies [becaa32]
  - ai@6.0.286
## @ai-sdk/azure@3.0.122

### Patch Changes

- Updated dependencies [e53d917]
  - @ai-sdk/openai@3.0.114
## @ai-sdk/langchain@2.0.294

### Patch Changes

- Updated dependencies [8a9722e]
- Updated dependencies [6a0f2d9]
- Updated dependencies [db02971]
- Updated dependencies [4d91999]
- Updated dependencies [b2ad994]
- Updated dependencies [9c85bd4]
- Updated dependencies [becaa32]
  - ai@6.0.286
## @ai-sdk/llamaindex@2.0.286

### Patch Changes

- Updated dependencies [8a9722e]
- Updated dependencies [6a0f2d9]
- Updated dependencies [db02971]
- Updated dependencies [4d91999]
- Updated dependencies [b2ad994]
- Updated dependencies [9c85bd4]
- Updated dependencies [becaa32]
  - ai@6.0.286
## @ai-sdk/mcp@1.0.82

### Patch Changes

- 1620182: fix(mcp): avoid duplicate legacy SSE OAuth refreshes for
stale 401 responses
## @ai-sdk/openai@3.0.114

### Patch Changes

- e53d917: fix(openai): preserve provider file references in Responses
tool results
## @ai-sdk/react@3.0.289

### Patch Changes

- Updated dependencies [8a9722e]
- Updated dependencies [6a0f2d9]
- Updated dependencies [db02971]
- Updated dependencies [4d91999]
- Updated dependencies [b2ad994]
- Updated dependencies [9c85bd4]
- Updated dependencies [becaa32]
  - ai@6.0.286
## @ai-sdk/rsc@2.0.286

### Patch Changes

- Updated dependencies [8a9722e]
- Updated dependencies [6a0f2d9]
- Updated dependencies [db02971]
- Updated dependencies [4d91999]
- Updated dependencies [b2ad994]
- Updated dependencies [9c85bd4]
- Updated dependencies [becaa32]
  - ai@6.0.286
## @ai-sdk/svelte@4.0.286

### Patch Changes

- Updated dependencies [8a9722e]
- Updated dependencies [6a0f2d9]
- Updated dependencies [db02971]
- Updated dependencies [4d91999]
- Updated dependencies [b2ad994]
- Updated dependencies [9c85bd4]
- Updated dependencies [becaa32]
  - ai@6.0.286
## @ai-sdk/vue@3.0.286

### Patch Changes

- Updated dependencies [8a9722e]
- Updated dependencies [6a0f2d9]
- Updated dependencies [db02971]
- Updated dependencies [4d91999]
- Updated dependencies [b2ad994]
- Updated dependencies [9c85bd4]
- Updated dependencies [becaa32]
  - ai@6.0.286

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>