Skip to main content

Roo Code 3.41.1 Release Notes (2026-01-16)

This patch release improves Orchestrator cost visibility, makes settings navigation faster, and fixes multiple provider compatibility issues.

Roo Code v3.41.1 Release

Aggregated Orchestrator subtask costs​

Orchestrator parent tasks now show a cumulative cost total that aggregates spend across the full subtask tree, making it easier to understand the true end-to-end cost of an orchestrated workflow at a glance. (thanks taltas!) (#10757)

Settings search now includes tab titles​

Settings search now indexes tab/section titles (like “MCP” or “Providers”), so you can search for a section name and jump directly to the right part of the settings UI instead of hunting through individual settings. (#10761)

QOL Improvements​

  • Clarifies that Slack and Linear integrations are only available on the Roo Code Cloud Team plan, reducing confusion when comparing plans. (#10748)

Bug Fixes​

  • Fixes an issue where Anthropic native tool calling could fail with a 400 error after stream retries/reconnections due to duplicate tool call IDs. (#10760)
  • Fixes an issue where OpenAI Native streaming could drop tool identity fields mid-stream, causing tool calls to be missed or fail in some OpenAI-compatible streams. (#10719)
  • Fixes an issue where the OpenAI Responses API could reject tool calls when call_id exceeded the 64-character limit, preventing 400 validation errors during tool use. (#10763)
  • Fixes an issue where Gemini requests (for example via OpenRouter) could fail with thought-signature validation errors mid-task, improving reliability for Gemini tool-calling chats. (#10694)
  • Fixes an issue where Gemini requests could fail when a user message contained an empty text block, ensuring only non-empty content is sent while preserving images. (#10728)
  • Fixes an issue where some MCP tool schemas could be rejected by providers when they used top-level anyOf/oneOf/allOf, improving cross-provider tool schema compatibility. (#10726)

Provider Updates​

  • Prevents Ollama setups from offering models that can’t use native tool calling, avoiding loops where the model outputs tool JSON as text instead of actually invoking tools. (#10735)