NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
npm
npm, #4234 most downloaded on npm
Latest stable
4.0.122
Last release 5 days ago
17 Nov 2026
Ships on a steady schedule
a new release about every 8 days
Most releases are documented
notes for 41 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
3 years old
751 releases · first in 2024
Drag across the ruler to compare, or tap a release to open it.
3bea041: fix(google): reject structured output schemas in batch requests because inline Google batch requests ignore responseJsonSchema
3bea041: fix(google): reject structured output schemas in batch requests because inline Google batch requests ignore responseJsonSchema
b0fccbd: fix(provider/google): redact MCP server credentials in telemetry
Google Interactions MCP server headers can contain connector credentials, which
were recorded in tool telemetry. google.tools.mcpServer now provides a
redacted telemetry view of its header values, and Interactions request bodies do
the same for provider error reporting, while the actual request keeps the
credentials.
a4de3a2: fix(google): include promptFeedback.blockReasonMessage in provider metadata
Updated dependencies [10ff0b1]
Updated dependencies [e878ab3]
Updated dependencies [d99098a]
f5fc6ba: Derive known model ID suggestions from each provider's model catalog declarations so catalog updates automatically update autocomplete, while
reasoning in generation and streaming calls. Map reasoning to model-supported thinking levels, including Hydrogen's xhigh, preserve explicit thinkingLevel overrides and API defaults, and warn when reasoning is unsupported.Your coding agent can read these notes before it upgrades. Set up the MCP server →
6d50947: add experimental extraction models backed by structured language model output
2ebc78b: Forward the provider's configured fetch transport when downloading file inputs. Add a fetch option to createDownload so core downloads can al
2ebc78b: Forward the provider's configured fetch transport when downloading file inputs. Add a fetch option to createDownload so core downloads can also use guarded-fetch for opt-in DNS validation on Node.js. The function returned by createDownload can now also be passed as experimental_download to generateText and streamText, and experimental_download now also handles generated file URLs returned by the model, so one hook covers all core downloads. Document that the default URL checks do not resolve DNS and recommend the guarded transport or network egress controls for untrusted URLs.
Keep Azure image and Vertex tool-result downloads separate from provider authentication wrappers so file hosts never receive Azure bearer tokens or Vertex Express Mode API keys.
Pass the call's abort signal to custom download requests, including prompt assets and generated files, so experimental_download: createDownload({ fetch }) preserves cancellation and timeouts.
4e43721: Add the gemini-4-hydrogen model ID and support Hydrogen's xhigh thinking level in Google Generate Content and Interactions provider options. Map portable reasoning: 'xhigh' to xhigh only for Hydrogen models.
Updated dependencies [2ebc78b]
Updated dependencies [cb4fa52]
f703361: use the native gemini decisions api for google decision models
6d50947: add experimental extraction models backed by structured language model output
gemini-4-carbon model with the same capabilities as gemini-4-argon.2ebc78b: Forward the provider's configured fetch transport when downloading file inputs. Add a fetch option to createDownload so core downloads can al
2ebc78b: Forward the provider's configured fetch transport when downloading file inputs. Add a fetch option to createDownload so core downloads can also use guarded-fetch for opt-in DNS validation on Node.js. The function returned by createDownload can now also be passed as experimental_download to generateText and streamText, and experimental_download now also handles generated file URLs returned by the model, so one hook covers all core downloads. Document that the default URL checks do not resolve DNS and recommend the guarded transport or network egress controls for untrusted URLs.
Keep Azure image and Vertex tool-result downloads separate from provider authentication wrappers so file hosts never receive Azure bearer tokens or Vertex Express Mode API keys.
Pass the call's abort signal to custom download requests, including prompt assets and generated files, so experimental_download: createDownload({ fetch }) preserves cancellation and timeouts.
6eff958: feat(openai): add positioned additionalTools and accept custom and namespace tools
a279e86: fix(openai): keep tool-result additionalTools for parallel results and openai options
Updated dependencies [2ebc78b]
Updated dependencies [cb4fa52]
x-gemini-service-tier response header when usage metadata omits it, for generated and streamed responses.6400cf8: Add resolveProviderOptions to parse provider options that a model accepts under several providerOptions keys, either as aliases (first) or me
resolveProviderOptions to parse provider options that a model accepts under several providerOptions keys, either as aliases (first) or merged in order (merge), and use it for the existing multi-key lookups in the OpenAI, OpenAI-compatible, Anthropic, Google, and Amazon Bedrock models. Resolution behavior for these models is unchanged.providerOptions.google.labels through to Google and Vertex AI Interactions requests, including streaming, so callers can attach request metadata such as safety_identifier.toolChoice is none so disabled tools cannot retrieve data or execute server-side. Preserve function declarations needed by conversation history while disabling function calling. This also applies when structured output uses a synthetic JSON response tool.27a8a77: Refactor Google Vertex model ID logic into a model catalog that declares capabilities and provider options, and generate provider capability
fd7e8a7: Preserve the order of text and complete function calls in streamed Google responses, including standalone thought signatures on empty text or
fd7e8a7: Preserve the order of text and complete function calls in streamed Google responses, including standalone thought signatures on empty text or reasoning parts.
Keep provider metadata on empty text parts when converting results to response messages so it can be passed back in subsequent requests.
Omit foreign-provider empty signature text when replaying history through Anthropic or Bedrock, preserving Bedrock’s existing native reasoning-history handling.
ef2dfb5: fix: pass through seed: 0 instead of dropping it in image and video models. KlingAI now reports seed: 0 as an unsupported setting like any other seed.
892af4d: feat (provider/google): implement server-side realtime extensions
GoogleRealtimeModel supports the existing getServerWebSocketConfig() contract, returning the unconstrained Gemini Live endpoint with resolved credential headers. Each server connection can create its own event parser. Reported gross token totals include thinking and tool-use counts, with cached input portions and available modality counts normalized without guessing missing values. Usage in a completion frame is attached to response-done; other snapshots use custom events with rawType: 'usageMetadata' and typed usage, preserving lifecycle ordering without inferring response associations.
3c6fb43: Forward explicit speech language codes to Gemini TTS for single- and multi-speaker requests, including the shared Vertex Gemini adapter. Omitted languages and language: 'auto' retain automatic detection without an unsupported-language warning.
Updated dependencies [36802d4]
Updated dependencies [e0143b0]
Updated dependencies [934100c]
Updated dependencies [892af4d]
Updated dependencies [89aff56]
Updated dependencies [41465fd]
Updated dependencies [9754014]
Updated dependencies [4bb73b0]
Updated dependencies [2e44144]
a32df11: Omit all Google Interactions tools when toolChoice is none, including provider-defined tools.
toolChoice is none, including provider-defined tools.4ae6986: fix(google): handle cycles and BigInt in functionResponse and parse string functionCall arguments
UNAVAILABLE and RESOURCE_EXHAUSTED are marked retryable.7716e18: Support promises and synchronous or asynchronous callbacks in provider API key and token settings. Resolve credentials per authenticated requ
7716e18: Support promises and synchronous or asynchronous callbacks in provider API key and token settings. Resolve credentials per authenticated request while preserving existing authentication behavior and synchronous credential serialization.
The experimental OpenAI Live getServerWebSocketConfig() method can now resolve asynchronously. Await its result before using the connection settings.
dfed8e8: Consolidate Google language model capabilities and request policies into a model catalog shared by runtime resolution and generated documentation, preserving existing Google and Vertex request behavior.
Updated dependencies [7716e18]
Updated dependencies [5c247e9]
Updated dependencies [dcd2f8d]
9f108f8: close string values in tool-call input streamed with streamFunctionCallArguments on their final chunk
9f108f8: close string values in tool-call input streamed with streamFunctionCallArguments on their final chunk
b1bcea1: Add videoMetadata support for file parts. Pass startOffset, endOffset, and fps through per-file providerOptions to control video clipping and frame rate with the Google and Google Vertex providers.
61a315e: fix: map the top-level reasoning option consistently across providers
Providers now resolve reasoning the same way:
providerOptions is sent unchanged. OpenAI no longer drops providerOptions.openai.reasoningEffort values that are not in its list for the model, and Fireworks no longer rewrites providerOptions.fireworks.reasoningEffort. Fireworks also sends top-level reasoning values directly instead of rounding them to 'high' at most, except for model families known to reject 'none' (GPT-OSS, GLM-5.3, Muse Spark) or 'minimal' (Nemotron, Inkling).reasoning maps to the closest level the model supports, rounding down, and rounds up only when there is no lower level. For example, 'xhigh' now maps to 'high' instead of 'max' on Anthropic and Bedrock models without xhigh effort, and 'medium' maps to 'low' instead of 'high' on DeepSeek, Kimi K3, and Mistral GLM-5.3 models.'none' on a model that cannot disable reasoning maps to the model's lowest level with a compatibility warning, instead of being dropped or sending a request the API rejects. This affects GPT-6 models without 'none', xAI grok-4.5 and later, Groq GPT-OSS, Kimi K3, Perplexity, QuiverAI, Bedrock OpenAI gpt-oss models, and Gemini 2.5 Pro (minimum thinking budget). Gemini 3 and Anthropic and Bedrock between_tools mappings now emit the compatibility warning too.mapReasoningToProviderEffort now accepts any reasoning value, returns an optional providerOptionsEffort unchanged, and rounds levels that are missing from the effort map to the closest supported level. mapReasoningToProviderBudget maps 'none' to the minimum reasoning budget, always returns a budget, and uses the default percentage for levels missing from budgetPercentages. OpenAICompatibleChatConfig adds an optional mapReasoning function for mapping the top-level reasoning option.
Updated dependencies [61a315e]
### Patch Changes - Updated dependencies [57086ff] - Updated dependencies [95348b3] - @ai-sdk/provider-utils@5.0.70
### Patch Changes - Updated dependencies [e64ed82] - @ai-sdk/provider-utils@5.0.69
f5fc6ba: Derive known model ID suggestions from each provider's model catalog declarations so catalog updates automatically update autocomplete, while
e729d58: chore(docs): update README.md for consistent format and to avoid duplication
bbe2bd1: fix(google): keep empty arrays and objects in tool-call input streamed with streamFunctionCallArguments
Vertex sends a value-less partialArgs entry for an empty container (e.g. $.tags[]), which was dropped, so [] and {} arguments went missing from the streamed tool-call input.
Updated dependencies [5475347]
Updated dependencies [e729d58]
6d50947: add experimental extraction models backed by structured language model output
### Patch Changes - Updated dependencies [2ebc78b] - Updated dependencies [cb4fa52] - @ai-sdk/provider-utils@5.0.80 - @ai-sdk/provider@4.0.38
43ad966: Report interrupted when a stream ends cleanly without its provider completion signal, preserving partial output without emitting an error solely for the interruption.
This applies to OpenAI (Chat, Responses, and Completions), Anthropic, Google (Generate Content and Interactions), Alibaba, Amazon Bedrock Converse, Cohere, DeepSeek, Groq, Hugging Face Responses, Mistral Chat, Moonshot AI, Open Responses, Perplexity, and xAI Responses. It also covers cutoffs in later Anthropic messages and Google automatic continuations.
Compatibility notice: These streams previously fell back to other or omitted the finish part. Applications must now handle interrupted in finish-reason switches, mappings, and custom validation. Check finishReason === 'interrupted' to detect incomplete responses; interruption alone does not trigger error handling.
Explicit unknown completion reasons still map to other. Actual provider, parsing, and transport errors retain their error paths; affected error paths that previously left the finish reason as other now report error. OpenAI-compatible remains unchanged.
Updated dependencies [10026d1]
39a776b: Add a shared encodePathSegment helper and use it to encode provider-returned identifiers in credentialed follow-up request paths.
### Patch Changes - Updated dependencies [0c2a9de] - Updated dependencies [200c47f] - @ai-sdk/provider@4.0.31 - @ai-sdk/provider-utils@5.0.64
d08bb66: Preserve grounding metadata from each automatically continued Google response in groundingMetadataPerRequest. Previously, a later response wi
groundingMetadataPerRequest. Previously, a later response without grounding could leave earlier Search queries absent from the final provider metadata even though their sources were returned. The new array preserves request order and includes null for requests without grounding.ae42699: fix(google): surface in-stream errors
4cb3dec: feat(provider/google): allow AUDIO response modalities for Google language models
### Patch Changes - Updated dependencies [130e871] - @ai-sdk/provider@4.0.28 - @ai-sdk/provider-utils@5.0.61
### Patch Changes - Updated dependencies [2e89fa4] - @ai-sdk/provider-utils@5.0.60
bdbeba3: feat(google): add gemini-4-argon model id and long decode support for generateContent: requests that reach the per-request cap (CONTINUATION)
gemini-4-argon model id and long decode support for generateContent: requests that reach the per-request cap (CONTINUATION) are continued automatically with usage summed across requests and the first request's prompt token count exposed as providerMetadata.google.initialPromptTokenCount, and the continuationToken provider option / provider metadata resume responses that used up their maxOutputTokens budget. generateText with the Interactions API now polls background: true model calls until they finish.2060ab9: fix(google): handle regional Gemini model IDs and valid thinking configurations
5cf13d2: feat: map the top level embedding dimensions to provider settings
### Patch Changes - Updated dependencies [9184a35] - @ai-sdk/provider@4.0.25 - @ai-sdk/provider-utils@5.0.57
32ff7ad: Preserve tool errors in OpenAI Chat Completions and Responses by wrapping their payloads in an error object. Send Google tool errors and deni
error object. Send Google tool errors and denied executions under functionResponse.response.error.### Patch Changes - Updated dependencies [d3bcad9] - @ai-sdk/provider@4.0.24 - @ai-sdk/provider-utils@5.0.56
fc1e19e: fix: limit jsonl rows to 64mb during batch result downloads
fc1e19e: fix: limit jsonl rows to 64mb during batch result downloads
8c0e699: feat(google): preserve file search custom metadata in retrieved context grounding chunks
753f2e1: feat(google): add the 'gemini-nano-banana-2.1' model ID
268225c: feat(provider): add a portable max reasoning level with native provider mappings, compatibility coercions, and budget-based fallback support.
Compatibility notes:
max widens LanguageModelV4CallOptions['reasoning'] and the derived ReasoningLevel type. Third-party providers and consumers with exhaustive switches or Record<ReasoningLevel, ...> mappings must handle max before upgrading. Ordinary calls using existing reasoning levels remain accepted.reasoning: 'xhigh' now send maxReasoningEffort: 'high' instead of the unsupported 'max', and return a compatibility warning. Anthropic and OpenAI Bedrock effort mappings retain their existing xhigh behavior. Partial Bedrock reasoning configurations continue to preserve explicit values and derive missing fields; explicit effort/budget overrides and disabled thinking no longer emit warnings for portable mappings that are not sent.minimal and xhigh calls keep their request mappings to low and high, respectively, and now return compatibility warnings in generation and streaming results. Portable max maps to high with the same warning behavior.Release classification: This remains a patch changeset under the repository's explicit release policy, which uses patch releases for both fixes and features. The TypeScript source-compatibility caveat for exhaustive consumers is disclosed above. Maintainers can override this classification with a major label if they decide to align the provider-spec addition with a future AI SDK major release.
Updated dependencies [fc1e19e]
Updated dependencies [5094ebd]
Updated dependencies [268225c]
Updated dependencies [0af2f7c]
2959d35: feat: rename evaluate to decide
evaluate to decidef5fc6ba: Derive known model ID suggestions from each provider's model catalog declarations so catalog updates automatically update autocomplete, while
### Patch Changes - Updated dependencies [6d50947] - @ai-sdk/provider-utils@5.0.81
### Patch Changes - Updated dependencies [2ebc78b] - Updated dependencies [cb4fa52] - @ai-sdk/provider-utils@5.0.80 - @ai-sdk/provider@4.0.38
### Patch Changes - Updated dependencies [6d50947] - @ai-sdk/provider-utils@5.0.81
### Patch Changes - Updated dependencies [2ebc78b] - Updated dependencies [cb4fa52] - @ai-sdk/provider-utils@5.0.80 - @ai-sdk/provider@4.0.38
32cf2f6: fix(google): send the default Gemini Live thinkingLevel when thinkingConfig sets neither thinkingLevel nor thinkingBudget, and stop overwriting a raw generationConfig.thinkingConfig
525efc5: feat(provider): advertise image model file and mask input support
Use confirmed model IDs for capability declarations so unrecognized model names remain unknown. Include Together AI FLUX.2 Pro and Flex single-image editing, and allow asynchronous capability lookups and middleware overrides to resolve to unknown.
Advertise QuiverAI Arrow 2 and Arrow 2 Telos file-input support, and mark Together AI Gemini image inputs unsupported by the current single-image request mapping.
Updated dependencies [e3605f6]
Updated dependencies [525efc5]
### Patch Changes - Updated dependencies [2ebc78b] - Updated dependencies [cb4fa52] - @ai-sdk/provider-utils@5.0.80 - @ai-sdk/provider@4.0.38
8beac3e: fix(google): serialize JSON Schema references in function responses
fe07867: Fix Google embedMany calls with more than 100 values by keeping per-value multimodal content aligned across automatic batches, including text
fe07867: Fix Google embedMany calls with more than 100 values by keeping per-value multimodal content aligned across automatic batches, including text-only entries. Validate content length before sending requests and validate each batch's provider options after middleware transforms them.
2db5621: fix(google): preserve code execution parts when replaying messages
2693319: Add Gemini 3.8 TTS support with structured speech metadata and per-turn speaker and style controls for prebuilt voices. Preserve native WAV responses without adding a second header, support explicit raw PCM, mu-law, and A-law output, and identify headerless audio formats correctly. Add the Gemini 3.8 speech model IDs to Google and Gateway types.
Share transcript and custom-voice inspection through the Google provider internal export, and reject empty speech transcripts before sending a request. Default newer and custom model IDs to structured speech while preserving the legacy format for Gemini 2.5 and 3.1.
771e74b: chore: enable dead code lint rules
Updated dependencies [fe07867]
Updated dependencies [a4b0940]
Updated dependencies [771e74b]
94d5d6d : Support reasoningEffortUpdate: 'none' for GPT-6 Sol and Luna in request-level options and positioned system messages. Validate update effort
reasoningEffortUpdate: 'none' for GPT-6 Sol and Luna in request-level options and positioned system messages. Validate update efforts against the model's supported efforts, warning and omitting unsupported request-level updates and rejecting unsupported historical updates.8dbe0be: fix(google): preserve image candidate finish reasons in provider metadata
### Patch Changes - Updated dependencies [2973485] - Updated dependencies [a4db5ea] - Updated dependencies [2937ea2] - @ai-sdk/provider-utils@5.0.45
e369c4d: fix(google): advertise the supported Gemini image per-call limit
d4d96bf: Add google.evaluationModel() for experimental Choice, Score, and Boolean evaluations through Gemini structured output, preserving provider th
google.evaluationModel() for experimental Choice, Score, and Boolean evaluations through Gemini structured output, preserving provider thinking options and validating exact labels and score bounds. Boolean answers contain prompted P(true) estimates validated to be in [0, 1]. Boolean estimates are not guaranteed to be calibrated; application code chooses thresholds.a22b5b2: fix(google): preserve prompt feedback and metadata across streaming chunks
4a994ad: feat(google): realtime session options for Gemini 3.8 Live
4a994ad: feat(google): realtime session options for Gemini 3.8 Live
Add thinkingConfig (thinkingLevel, thinkingBudget, includeThoughts) and
defaultToolBehavior to GoogleRealtimeModelOptions. thinkingConfig is merged
into the Live setup.generationConfig. Background-reasoning Live models such as
gemini-3.8-live-extended-thinking require exactly one of thinkingLevel or
thinkingBudget, so the provider sends thinkingLevel: 'low' on those models
when neither is set. defaultToolBehavior stamps behavior on every function
declaration in the setup.
Forward the Live interactionStatus and waitingForInput server messages as
custom events so applications can tell when a background-reasoning model is idle,
since turnComplete alone no longer means that.
### Patch Changes - Updated dependencies [5c0054d] - Updated dependencies [39535af] - @ai-sdk/provider@4.0.15 - @ai-sdk/provider-utils@5.0.41
2a32459: fix(google): preserve JSON Schema instead of converting to OpenAPI schema
2a32459: fix(google): preserve JSON Schema instead of converting to OpenAPI schema
d93e295: fix(google): keep a realtime functionResponse.response an object
Gemini types that field as a google.protobuf.Struct, which accepts an object and
nothing else. onToolCall returns unknown and addToolOutput takes unknown, so a
string, number, array or null tool result reached the wire unwrapped, Gemini closed
the socket with 1007, and the close code was dropped on the way back so the
application saw a plain disconnect. A non-object result is now wrapped under the
output key the field's own docstring prescribes; an object is passed through
unchanged, as before.
An output that is not valid JSON no longer becomes {}. That branch told the model
the tool had returned an empty object, with nothing thrown and the socket still up, so
the answer was wrong with nothing to notice. The text is kept instead.
f88c7dc: fix(vertex): download tool result file URLs
5ec21a6: fix: reject unsupported batch request types
ab6e9f9: feat(google): add batch cancellation and listing
### Patch Changes - Updated dependencies [912fb01] - @ai-sdk/provider@4.0.12 - @ai-sdk/provider-utils@5.0.38
a4ba394: feat: support per-request models in batch
d4485fe: feat(ai): support minItems and maxItems in array outputs
18ad19c: feat(google): support agentic video processing in the Interactions API
providerMetadata.<provider>.inputFileId / inputFileExpiresAt) and accept an inputFileExpiresAfter provider option on the OpenAI and xAI batch input file upload### Patch Changes - Updated dependencies [6bcc0f8] - @ai-sdk/provider-utils@5.0.36
5190b67: feat(provider): extend the FilesV4 interface with optional getFileMetadata, downloadFile (streaming), and deleteFile operations, plus abortSi
getFileMetadata, downloadFile (streaming), and deleteFile operations, plus abortSignal/headers call options and a { type: 'stream' } upload data variant; upload results now expose byteSize, createdAt, and expiresAt (also surfaced by the core uploadFile() helper, which now forwards abortSignal/headers); add postMultipartStreamToApi (streaming multipart uploads with deterministic part ordering and failure-path stream teardown), deleteFromApi, and createBinaryStreamResponseHandler to provider-utilsOne column per month.