165d531: deprecate searchParameters (xai live search) in favor of web_search/x_search agent tools
4d53460: feat(provider/xai): add enableImageSearch to the xAI Web Search tool

The xAI Responses API supports enable_image_search on Web Search tools. xai.tools.webSearch() now accepts enableImageSearch and sends it through to the API as enable_image_search.

Ping requests now respond per JSON-RPC spec

↗

@ai-sdk/mcp@1.0.45

Patch Changes

ec5fceb: fix(mcp): respond to ping requests with an empty result per JSON-RPC spec (closes #6282)

May 29, 2026

Web search tool added to Responses API

↗+1

@ai-sdk/azure@3.0.68

Patch Changes

c679fec: feat(provider/azure):web search tool in the Azure OpenAI Responses API.
Updated dependencies [c679fec]
- @ai-sdk/openai@3.0.67

May 28, 2026

Tool input no longer re-validated for output-error parts

↗+7

ai@6.0.193

Patch Changes

af580ea: fix(ai): do not re-validate tool input for output-error parts in validateUIMessages

Claude Opus 4.8 now supported

↗+11

@ai-sdk/gateway@2.0.94

Patch Changes

0e2aaaa: feat(provider/anthropic): add support for claude-opus-4-8
8fc77c6: Backport: chore(provider/gateway): update gateway model settings files

Dependency update

↗+12

@ai-sdk/google-vertex@4.0.140

Patch Changes

4084fcd: feat(provider/anthropic): add support for claude-opus-4-8
Updated dependencies [4084fcd]
- @ai-sdk/anthropic@3.0.81

May 27, 2026

Web search queries now forwarded from Responses API

↗+1

@ai-sdk/openai@3.0.66

Patch Changes

c82ab42: feat(openai): forward web_search_call.action.queries from Responses API

Document files now supported in tool results

↗

@ai-sdk/amazon-bedrock@4.0.110

Patch Changes

20971fc: fix(amazon-bedrock): support document files in tool results

MCPClientError now carries HTTP context—statusCode, url, responseBody

↗

@ai-sdk/mcp@1.0.44

Patch Changes

77775a4: feat(mcp): expose statusCode, url, and responseBody on MCPClientError for HTTP transport failures

MCPClientError now carries structured HTTP context when it originates from the streamable HTTP transport. This lets downstream consumers (e.g. agent frameworks that need to decide whether to fall back from streamable HTTP to legacy SSE transport per the MCP spec) branch on the actual response status without parsing the error message string.

Fields are optional — they remain undefined for stdio transport errors and for non-response failures (network errors, aborts).

Python AIMessageChunk now recognized in RemoteGraph streams

↗

@ai-sdk/langchain@2.0.198

Patch Changes

63847a6: fix(langchain): recognize Python AIMessageChunk plain message objects from RemoteGraph streams

Python langchain-core serializes streaming message chunks with type: "AIMessageChunk", while TypeScript langchain-core uses type: "ai". The toUIMessageStream adapter previously only matched the TypeScript form, silently dropping text deltas and tool-call events when streaming from Python LangGraph servers via RemoteGraph.

Gemini 3 tool calls auto-injected with signature validator on replay

↗+1

@ai-sdk/google@3.0.80

Patch Changes

f62ffe0: fix(google): auto-inject skip_thought_signature_validator for Gemini 3 tool-call replays without a signature

Gemini 3 models reject requests when an assistant functionCall part lacks a thoughtSignature with HTTP 400 "Function call is missing a thought_signature in functionCall parts." This is easy to hit when application code persists/serializes messages and drops providerOptions.google.thoughtSignature (custom DB schemas, useChat server routes that rebuild messages, synthetic tool-call injection).

The provider now detects this case (Gemini 3 model + missing signature under google, googleVertex, and vertex namespaces) and injects the documented skip_thought_signature_validator sentinel into the outbound functionCall, plus surfaces a one-shot warning per request listing the affected tool names so the developer can find and fix the upstream serialization. Non-Gemini-3 models are unaffected, and real signatures take precedence when present.