fix(max-tokens): anthropic models streaming vs non-streaming

[icecrasher321]

Summary

Anthropic models streaming vs non-streaming token limits.

Type of Change

• Bug fix
• New feature

Testing

Tested manually with @Sg312

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Review	Updated (UTC)
docs	Ready	Preview, Comment	Jan 27, 2026 0:49am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR implements differentiated token limits for Anthropic models based on whether requests are streaming or non-streaming, addressing timeout issues with non-streaming requests.

Key Changes:

• Added maxOutputTokens capability to model definitions with separate max (streaming) and default (non-streaming) values
• Anthropic Claude 4.x/Opus models now use 64K tokens for streaming, 4096 for non-streaming (previously hardcoded 1024)
• Claude 3.x computer use models use 8192 tokens for both modes
• Implemented getMaxOutputTokensForModel() function to dynamically select appropriate limits
• Added UI field for users to manually configure max output tokens
• Increased browser automation timeout from 3 to 10 minutes (related change)

Issues Found:

• Documentation inconsistency in apps/sim/providers/utils.ts:999 - comment states 8192 but implementation uses 4096

Confidence Score: 4/5

• Safe to merge with one minor documentation fix needed
• The implementation is sound and addresses a real issue with timeout problems on non-streaming requests. The logic correctly differentiates between streaming (max tokens) and non-streaming (default tokens) modes. Only issue is a documentation comment that doesn't match the implementation (states 8192 but code uses 4096).
• Pay attention to apps/sim/providers/utils.ts for the documentation fix

Important Files Changed

Filename	Overview
apps/sim/providers/models.ts	Added maxOutputTokens capability with streaming vs non-streaming limits for Anthropic models; implemented getMaxOutputTokensForModel() function
apps/sim/providers/anthropic/index.ts	Updated default max_tokens from hardcoded 1024 to dynamic value using getMaxOutputTokensForModel() based on streaming mode
apps/sim/providers/bedrock/index.ts	Updated default maxTokens from hardcoded 4096 to dynamic value using getMaxOutputTokensForModel() based on streaming mode
apps/sim/providers/utils.ts	Added getMaxOutputTokensForModel() wrapper function with documentation inconsistency (mentions 8192 but returns 4096)
apps/sim/blocks/blocks/agent.ts	Added maxTokens UI input field and schema definition to allow users to configure max output tokens
apps/sim/tools/browser_use/run_task.ts	Increased MAX_POLL_TIME_MS from 180000 (3 min) to 600000 (10 min) to accommodate longer browser automation tasks

Sequence Diagram

sequenceDiagram
    participant User
    participant Agent
    participant Provider as Anthropic/Bedrock Provider
    participant Models as models.ts
    participant API as Anthropic/Bedrock API

    User->>Agent: Configure agent with model & maxTokens
    Agent->>Provider: executeRequest(model, stream, maxTokens)
    
    alt maxTokens provided by user
        Provider->>Provider: Use Number.parseInt(request.maxTokens)
    else maxTokens not provided
        Provider->>Models: getMaxOutputTokensForModel(model, stream)
        
        alt Streaming request
            Models-->>Provider: Return maxOutputTokens.max (e.g., 64000)
        else Non-streaming request
            Models-->>Provider: Return maxOutputTokens.default (e.g., 4096)
        end
    end
    
    Provider->>API: API call with determined max_tokens
    API-->>Provider: Response (within token limit)
    Provider-->>Agent: ProviderResponse
    Agent-->>User: Generated content

Loading

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[icecrasher321]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR