NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
npm · #4234 most downloaded on npm
Last release today
18 Sep 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Nothing withdrawn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
2 years old
669 releases · first in 2024
2565e70: feat(google): add support for image search, replace obsolete google_search_retrieval implementation
### Patch Changes - Updated dependencies [31205a4] - @ai-sdk/provider-utils@4.0.46
One column per month.
Updated dependencies [ 9e1d1b2 ]
gemini-3.1-flash-lite-preview### Patch Changes - Updated dependencies [08336f1] - @ai-sdk/provider-utils@4.0.17
9435df8: feat(provider/moonshotai): normalize tool schemas for Moonshot's MFJS validator. Tuple items arrays become prefixItems, type next to anyOf mo
items arrays become prefixItems, type next to anyOf moves into the branches, and non-object root schemas fail with a clear client-side error instead of Moonshot's opaque 400. Everything else passes through unchanged; the original schema is still used for AI SDK result validation.ba40fab: fix(fireworks): enable structured outputs so response_format json_schema is forwarded instead of degrading to json_object
@ai-sdk/openai-compatible@3.0.31
@ai-sdk/openai-compatible@3.0.31
86892f3: fix(openai-compatible): preserve unmapped usage fields in usage.raw
86892f3: fix(openai-compatible): preserve unmapped usage fields in usage.raw
usage.raw is specified as usage "in the shape that the provider returns",
and the chat model already parsed the usage object loosely so that extra
top-level fields survived. The nested prompt_tokens_details and
completion_tokens_details objects were still strict, so anything a provider
reported inside them was dropped — which is where providers put their most
distinguishing detail. The existing "should preserve extra usage fields"
fixture was itself losing audio_tokens, image_tokens and text_tokens
this way.
Both nested objects are now parsed loosely, as is the completion model's usage
schema, which was strict throughout despite feeding the same raw field.
Only usage.raw changes. The mapped token counts are unaffected.
765b013: feat(provider/google): add support for gemini-3.1-pro-preview
gemini-3.1-pro-preview### Patch Changes - Updated dependencies [4024a3a] - @ai-sdk/provider-utils@4.0.15
5a307f5: feat(provider/google-vertex): allow using Gemini image models with generateImage
generateImage051361b: fix(vertex): add fallback for providerOptions keyname
4c27179: feat(google): allow using Gemini image models with generateImage
generateImage99fbed8: feat: normalize provider specific model options type names and ensure they are exported
3b3e32f: fix(google): handle thoughtSignature on empty-text parts in doGenerate and doStream
ba98c56: fix(google): make codeExecutionResult.output optional in response schema
ba98c56: fix(google): make codeExecutionResult.output optional in response schema
Gemini 3 Flash omits the output field in codeExecutionResult when code execution produces no text output (e.g., only saves files). The Zod response schema now accepts a missing output field and defaults it to an empty string.
7168375: feat (ai, provider): default global provider video model resolution
ae30443: fix(google): remove shut down gemini-2.5-flash-image-preview
gemini-2.5-flash-image-preview53f6731: feat (ai, provider): experimental generate video support
### Patch Changes - Updated dependencies [96936e5] - @ai-sdk/provider-utils@4.0.12
### Patch Changes - Updated dependencies [2810850] - @ai-sdk/provider-utils@4.0.11 - @ai-sdk/provider@3.0.6
1524271: chore: add skill information to README files
97b3ebb: fix (provider/google): make segment optional in groundingSupports schema
97b3ebb: fix (provider/google): make segment optional in groundingSupports schema
The Google Generative AI API sometimes returns grounding supports without a segment field. This change makes the segment field optional to handle these responses correctly.
Updated dependencies [ 2973485 ]
Updated dependencies [ 0455398 ]
Updated dependencies [ 215b25e ]
Updated dependencies [ 11b4c2d ]
Updated dependencies [ 5c0054d ]
### Patch Changes - Updated dependencies [5c090e7] - @ai-sdk/provider@3.0.4 - @ai-sdk/provider-utils@4.0.8
Updated dependencies [ 9942196 ]
Updated dependencies [ 912fb01 ]
Updated dependencies [ a4ba394 ]
Updated dependencies [ e5a22f0 ]
Updated dependencies [ 6bcc0f8 ]
### Patch Changes - Updated dependencies [d937c8f] - @ai-sdk/provider@3.0.2 - @ai-sdk/provider-utils@4.0.4
### Patch Changes - Updated dependencies [0b429d4] - @ai-sdk/provider-utils@4.0.3
7fb07aa : fix(typesafe-ai): surface the error_type code instead of a generic failure message
@latestUpdated dependencies [ 3e125ba ]
366f50b: chore(provider): add deprecated textEmbeddingModel and textEmbedding aliases
0c3b58b: fix(provider): add specificationVersion to ProviderV3
0adc679: feat(provider): shared spec v3
9be07c8: feat(google): thinking_level option for Gemini 3
fd788ce: fix(provider/google): preserve nested empty object schemas and descriptions in tool parameters
8d9e8ad: chore(provider): remove generics from EmbeddingModelV3
Before
model.textEmbeddingModel("my-model-id");
After
model.embeddingModel("my-model-id");
2625a04: feat(openai); update spec for mcp approval
7728ac5: The mediaResolution option has been added and is now passed to the Google API.
9a728c8: support latest gemini model id
32a6c13: Add Google Maps grounding tool support for location-aware Gemini responses
95f65c2: chore: use import * from zod/v4
e300a3b: Fixed Zod validation error when using google.tools.fileSearch(). The Google File Search API returns fileSearchStore instead of uri in retrievedContext. Updated extractSources() function to handle both the old format (Google Search with uri) and new format (File Search with fileSearchStore), maintaining backward compatibility while preventing validation errors. Also fixed title handling to use undefined for URL sources and 'Unknown Document' for document sources.
218bba1: fix(google): use dynamic providerOptionsName when retrieving thoughtSignature in convertToGoogleGenerativeAIMessages
When using @ai-sdk/google-vertex provider with Gemini thinking models, multi-step tool calls would fail with "function call is missing a thought_signature" error. This was because thoughtSignature was stored under providerOptions.vertex but retrieved using hardcoded providerOptions.google. This fix passes providerOptionsName to convertToGoogleGenerativeAIMessages and uses it dynamically.
954c356: feat(openai): allow custom names for provider-defined tools
0b92881: Add Google Vertex RAG Engine grounding provider tool
544d4e8: chore(specification): rename v3 provider defined tool to provider tool
0c4822d: feat: EmbeddingModelV3
f8c981f: Fix adding google search along with url context in vertex ai
6078060: fix(provider/google): remove includethoughts warning
bb28cac: Change streamText loop to merge file part processing into main parts loop
fff8d59: feat(provider/google): Add support for the imageSize provider option
1742445: Support for custom provider name in google and anthropic providers
e8109d3: feat: tool execution approval
ed329cb: feat: Provider-V3
3bd2689: feat: extended token usage
1cad0ab: feat: add provider version to user-agent header
9b17031: Improve error message when mixing function tools with provider-defined tools to clarify fallback behavior and list ignored function tools
8dac895: feat: LanguageModelV3
ee50cc5: fix(provider/google): lazy schema loading
import time improved by 12.5% (22.3ms ➡️ 19.5ms)
457318b: chore(provider,ai): switch to SharedV3Warning and unified warnings
0ad470b: feat(provider/google): add enterpriseWebSearch tool
db913bd: fix(google): add thought signature to gemini 3 pro image parts
9061dc0: feat: image editing
8370068: fix(provider/google): preserve thoughtSignature through tool execution
ee8cd23: fix(vertex): allow 'vertex' as a key for providerOptions
7dea60e: add promptFeedback outputs
366f50b: chore(provider): add deprecated textEmbeddingModel and textEmbedding aliases
8ee8edc: Prepare search tool for gemini-3-pro-preview
2825757: Add Google File search tool
4616b86: chore: update zod peer depenedency version
33d9327: add gemini-3-pro-preview and gemini-3-pro-image-preview model IDs
0cfae4c: feat(vertex): support 'trafficType' in provider usageMetadata
09ba2dd: Support imageConfig.aspectRatio configuration for Gemini models
32a8c82: feat: add gemini 3 pro
166b6d7: fix(provider/google): preserve nested empty object schemas in tool parameters to fix "property is not defined" validation errors when using required properties with empty object types
522f6b8: feat: ImageModelV3
599a97f: fix: update gemini 3 model id
49e2b6a: fix(google): return request as object
3794514: feat: flexible tool output content support
cbf52cd: feat: expose raw finish reason
870297d: feat(google): gemini-3-flash
10c1322: fix: moved dependency @ai-sdk/test-server to devDependencies
c8003fb: fix(@ai-sdk/google): Make title field optional in grounding metadata schema
4d2e88e: fix(google,google-vertex): update known model IDs
e833473: chore (provider/google): Add preview modelIds for gemini 2.5 flash and lite
Updated dependencies
218bba1: fix(google): use dynamic providerOptionsName when retrieving thoughtSignature in convertToGoogleGenerativeAIMessages
218bba1: fix(google): use dynamic providerOptionsName when retrieving thoughtSignature in convertToGoogleGenerativeAIMessages
When using @ai-sdk/google-vertex provider with Gemini thinking models, multi-step tool calls would fail with "function call is missing a thought_signature" error. This was because thoughtSignature was stored under providerOptions.vertex but retrieved using hardcoded providerOptions.google. This fix passes providerOptionsName to convertToGoogleGenerativeAIMessages and uses it dynamically.
@ai-sdk/provider-utils@4.0.0-beta.59
2625a04: feat(openai); update spec for mcp approval
cbf52cd: feat: expose raw finish reason
@ai-sdk/provider-utils@4.0.0-beta.56
@ai-sdk/provider-utils@4.0.0-beta.55
fd788ce: fix(provider/google): preserve nested empty object schemas and descriptions in tool parameters
166b6d7: fix(provider/google): preserve nested empty object schemas in tool parameters to fix "property is not defined" validation errors when using r
@ai-sdk/provider-utils@4.0.0-beta.54
0ad470b: feat(provider/google): add enterpriseWebSearch tool
870297d: feat(google): gemini-3-flash
@ai-sdk/provider-utils@4.0.0-beta.52
32a6c13: Add Google Maps grounding tool support for location-aware Gemini responses
@ai-sdk/provider-utils@4.0.0-beta.51
@ai-sdk/provider-utils@4.0.0-beta.50
@ai-sdk/provider-utils@4.0.0-beta.49
@ai-sdk/provider-utils@4.0.0-beta.48
ee8cd23: fix(vertex): allow 'vertex' as a key for providerOptions
49e2b6a: fix(google): return request as object
@ai-sdk/provider-utils@4.0.0-beta.47
Your coding agent can read these notes before it upgrades. Set up the MCP server →