Fix structuredOutputMode for newer Anthropic models

[miguelg719]

why

New output mode output_format from anthropic breaks compat with older jsonTool. Opus 4.7 defaults to the new format so it breaks with strict mode settings enabled

what changed

On aisdk.ts ensured structuredOutputMode defaults to auto.

test plan

• tested with env:LOCAL and extract with custom schema

---

Summary by cubic

Set providerOptions.anthropic.structuredOutputMode to auto for the anthropic provider to align with newer Anthropic models. Restores structured outputs and prevents schema/response errors when generating JSON, and adds a patch changeset for @browserbasehq/stagehand.

^{Written for commit 6ef2148. Summary will update on new commits.}

[changeset-bot]

🦋 Changeset detected

Latest commit: 6ef2148

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 4 packages

Name	Type
@browserbasehq/stagehand	Patch
@browserbasehq/stagehand-evals	Patch
@browserbasehq/stagehand-server-v3	Patch
@browserbasehq/stagehand-server-v4	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[cubic-dev-ai]

1 issue found across 2 files

Confidence score: 3/5

• There is a concrete regression risk in packages/core/lib/v3/llm/aisdk.ts: structuredOutputMode: "auto" is gated to prefixed Anthropic model IDs, so unprefixed claude-* IDs may not get the structured-output fix.
• Because this is a medium-severity, high-confidence behavior gap (6/10, 8/10) in model routing logic, merge risk is moderate rather than low.
• Pay close attention to packages/core/lib/v3/llm/aisdk.ts - ensure Anthropic ID matching covers both prefixed and unprefixed forms so structured output behavior is consistent.

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/core/lib/v3/llm/aisdk.ts">

<violation number="1" location="packages/core/lib/v3/llm/aisdk.ts:182">
P2: `structuredOutputMode: "auto"` is only applied for prefixed Anthropic model IDs, so unprefixed Anthropic IDs (e.g. `claude-*`) still miss the new structured-output fix.</violation>
</file>

Architecture diagram

sequenceDiagram
    participant App as Stagehand App
    participant LLMClient as LLMClient
    participant AISdkClient as AISdkClient
    participant Provider as Provider SDK (Anthropic/Azure)
    participant AnthropicAPI as Anthropic API

    Note over App,AnthropicAPI: Structured output configuration flow

    App->>LLMClient: execute() with structured output schema
    LLMClient->>AISdkClient: callModel() with provider options

    alt Provider is "anthropic"
        AISdkClient->>AISdkClient: Set providerOptions.anthropic.structuredOutputMode = "auto"
        AISdkClient->>Provider: call with providerOptions including structuredOutputMode: "auto"
        Provider->>AnthropicAPI: API request with tool-use mode (structured output)
        AnthropicAPI-->>Provider: Structured JSON response
        Provider-->>AISdkClient: Parsed structured output
    else Provider is "azure"
        AISdkClient->>AISdkClient: Set providerOptions.azure.strictJsonSchema = true
        AISdkClient->>Provider: call with strictJsonSchema enabled
        Provider-->>AISdkClient: Structured output via strict schema
    else Default (no specific provider options)
        AISdkClient->>Provider: call without structured output overrides
        Provider-->>AISdkClient: Unstructured response (may fail for JSON)
    end

    AISdkClient-->>LLMClient: Return structured result
    LLMClient-->>App: Structured output data

Loading

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review, or fix all with cubic.}