Cline v4.1.26 adds separate cloud task model setting and restores Claude Sonnet 5.5 as default

Cline v4.1.26 adds a Local | Cloud toggle in Settings > API Configuration that lets cloud tasks run on a different Cline model than local tasks. The toggle appears when cloud sessions are on. The release also returns Claude Sonnet 5.5 as the Cline provider's default and refreshes the model catalog.

Cloud tasks

• The Cloud setting picks the Cline model that new cloud tasks run on. It is saved separately from local model choices. Clicking the model label in a cloud session's composer opens the setting.
• Starting or resuming a cloud sandbox now shows a progress card with the repo, branch and steps (start sandbox, clone repository, start agent). The card collapses once the agent starts.
• Cloud sessions now hide the Plan/Act toggle and the auto-approve bar instead of showing them disabled.

Model and provider changes

• With no model chosen, the Cline provider uses Claude Sonnet 5.5 again. It is back at the top of Cline's recommended list.
• Model requests retry at most 3 times instead of 5.
• Kimi K3 now receives max reasoning for Extra High and Max, replacing an xhigh setting the model rejects.
• The model catalog was refreshed. New providers: CommandCode, Reflection, Liquid Inference and Munito AI. Removed: CrofAI, iFlow and Llama.
• The Cline free list adds Reflection Beam, Nemotron 3.5 Lightning and Laguna S 2.1. It drops Mimo V2.6 Flash and Muse Spark 1.3 Contributor.
• ClinePass now defaults to Kimi K3.1 Pro and adds MiniMax M3.1 Flash and DeepSeek V4.1 Flash 1017.
• Default models change for most other providers, mostly to Kimi K3.1 Pro, MiniMax M3.1 Flash or DeepSeek V4.1 Flash 1017. Users who do not pin a model for a provider should expect a different default.

Fixes

• GPT-6 compatibility: GPT-6 models (such as gpt-6-sol and gpt-6.1-sol) now work on OpenAI and Azure OpenAI-compatible endpoints. SAP AI Core no longer rejects requests with "Unsupported parameter: 'max_tokens'". Both now get max_completion_tokens.
• Gemini API keys: The gemini provider no longer sends native Google Search. Free-tier keys failed every turn with a 429 quota error on Gemini 3+, and on Gemini 2.x every Cline tool was dropped.
• Plan/Act: Toggling while a task is starting no longer drops the first prompt, and the session no longer keeps running with the other mode's tools.
• Rules: Disabled rules stay disabled.
• Terminal focus: Starting a command no longer moves keyboard focus into the terminal, so text typed in chat no longer lands in the shell.
• Questions: Cancelling a task or opening another one while Cline is asking a question no longer lets the next message be taken as the answer to the old question. With ask_question, the model keeps going after an option is picked, and empty answers are rejected.
• Retries: A turn whose connection drops after the model has streamed only reasoning is now retried instead of ending the task.
• Auto-compaction: It no longer runs for models with an unknown context limit, and no longer runs again every few tool calls after the first compaction.
• run_commands: An empty command list now returns an error instead of a silent success. On PowerShell, the tool tells the model how quote escaping works and to confirm recursive deletes first.
• UI: Task header badges wrap instead of hiding the close button, the GitHub connection check no longer flashes a loading indicator, and the Local/Cloud toggle's padding is fixed.

The full changelog is on GitHub as the comparison between v4.1.25 and v4.1.26.

What this means

Separate Local and Cloud model settings mean a cloud task's model no longer has to match the one used locally. Hiding the Plan/Act toggle and auto-approve bar in cloud sessions removes controls that were not usable there.

The default changes matter most for users who rely on unpinned providers: the model behind a provider can change after this update, so pinning a model keeps behavior stable. The GPT-6 and Gemini fixes address request failures on specific endpoints and key types, and the auto-compaction, retry and ask_question fixes target tasks that ended early or behaved incorrectly.