Skip to main content

Roo Code 3.25.12 Release Notes (2025-08-12)

This release brings a massive context window upgrade for Claude Sonnet 4, configurable timeouts for local providers, and minimal reasoning support for OpenRouter.

Claude Sonnet 4: 1 Million Token Context Window​

We've upgraded our Claude Sonnet 4 integration to support Anthropic's latest API update, increasing the context window from 200,000 tokens to 1 million tokens - a 5x increase (#7005):

  • Massive Context: Work with entire codebases, extensive documentation, or multiple large files in a single conversation
  • Tiered Pricing Support: Automatically handles Anthropic's tiered pricing for extended context usage
  • UI Integration: Context window size now displays in the model info view with a toggle to enable the 1M context feature

This significant upgrade enables you to tackle much larger projects and maintain context across extensive codebases without splitting conversations.

📚 Documentation: See Anthropic Provider Guide for setup and usage instructions.

Configurable API Timeout for Local Providers​

Local AI providers running large models can now configure custom timeout settings to prevent premature disconnections (#6531):

  • Flexible Timeouts: Set timeouts from 0 to 3600 seconds (default: 600 seconds)
  • Provider Support: Works with LM Studio, Ollama, and OpenAI-compatible providers

Configure in your VSCode settings:

{
  "roo-cline.apiRequestTimeout": 1200  // 20 minutes for very large models
}

OpenRouter Minimal Reasoning Support​

OpenRouter now supports minimal reasoning effort for compatible models (#6998):

  • New Reasoning Level: 'Minimal' option available for specific models
  • UI Updates: Thinking budget interface shows minimal option when applicable
  • Optimized Performance: Better control over reasoning intensity for different tasks

This addition provides more granular control over model reasoning, allowing you to optimize for speed or depth based on your needs.

QOL Improvements​

  • GPT-5 Model Optimization: GPT-5 models excluded from 20% context window output token cap for better performance (#6963)
  • Task Management: Added expand/collapse translations for better task organization (#6962)
  • Localization: Improved Traditional Chinese locale with better translations (thanks PeterDaveHello!) (#6946)

Bug Fixes​

  • File Indexing: JSON files now properly respect .rooignore settings during indexing (#6691)
  • Tool Usage: Fixed tool repetition detector to allow first tool call when limit is 1 (#6836)
  • Service Initialization: Improved checkpoint service initialization handling (thanks NaccOll!) (#6860)
  • Browser Compatibility: Added --no-sandbox flag to browser launch options for better compatibility (thanks QuinsZouls!) (#6686)
  • UI Display: Fixed long model names truncation in model selector to prevent overflow (#6985)
  • Error Handling: Improved bridge config fetch error handling (#6961)

Provider Updates​

  • Amazon Bedrock: Added OpenAI GPT-OSS models to the dropdown selection (#6783)
  • Chutes Provider: Added support for new Chutes provider models (#6699)
  • Requesty Integration: Added Requesty base URL support (#6992)
  • Cloud Service: Updated to versions 0.9.0 and 0.10.0 with improved stability (#6964, #6968)
  • Bridge Service: Switched to UnifiedBridgeService for better integration (#6976)
  • Roomote Control: Restored roomote control functionality (#6796)