Skip to main content

Roo Code 3.41.0 Release Notes (2026-01-15)

This release adds OpenAI Codex subscription access, expands OpenAI model choices, and improves stability.

Roo Code v3.41.0 Release

OpenAI Codex provider with OAuth subscription access​

Adds the OpenAI Codex provider with subscription-based OAuth sign-in, so you can connect a ChatGPT Plus/Pro account to use Codex models without API keys or per-token billing. Just select OpenAI - ChatGPT Plus/Pro in the provider settings! (#10736)

GPT-5.2-Codex model option for OpenAI (Native)​

Adds the GPT-5.2-Codex model to the OpenAI (Native) provider so you can select the coding-optimized model with its expanded context window and reasoning effort controls. (#10731)

Bug Fixes​

  • Gemini sessions no longer fail after a provider switch: Resolves a streaming error where LiteLLM Gemini tool calls could fail with corrupted thought signatures when switching models mid-task. (#10743)
  • Long terminal runs no longer degrade memory: Fixes a memory leak where large command outputs could keep growing buffers after completion, leading to gray screens during long sessions. (#7666)

Misc Improvements​

  • End-to-end tests run reliably again: Restores MCP and subtask coverage and fixes flaky tool tests so contributors can run CI-like checks locally and catch regressions earlier. (thanks ArchimedesCrypto, dcbartlett!) (#10720)
  • Automated tests no longer stall on tool approvals: Fixes a problem where MCP end-to-end tests could hang on manual approval prompts by auto-approving time server tools. (thanks ArchimedesCrypto!) (#10733)