Replies that stop short
Cline asks the model to finish a reply that stopped short
Once, when a response ends without a recognized finish reason
When a model's reply ends without a finish reason Cline recognizes, Cline now asks the model to continue, once. Before this release, Cline treated a reply like that as complete.
Custom Anthropic base URLs
Claude through Azure AI Foundry and corporate gateways works again
The 400 error from the previous release is gone
If you reach Claude through a custom Anthropic base URL, such as Azure AI Foundry or a corporate gateway, the 400 errors stop. The previous release sent an option that only Anthropic's own API accepts.
Long MCP tool output
Cline can page through MCP output that overflows its context
It gets a preview, plus a link it pages through with read_files
When an MCP tool returns more than fits in context, Cline gets a preview and a link it can page through with its read_files tool. The output past the cutoff used to be lost.
Commit messages
Generate Commit Message now follows your .clinerules
Language, format and ticket references apply to commits too
Generate Commit Message now follows your .clinerules. The conventions you already set there, like the language, the format and ticket references, now shape your commit messages as well.
New default models
Nineteen providers switch to a new default model
Vultr renamed its model ids, so a pinned Vultr model may need picking again
Nineteen providers get a new default model, including Mistral, Nvidia and DigitalOcean. If you haven't pinned a model there, expect a different one. Vultr renamed its model ids upstream, so a pinned Vultr model may need selecting again.
Recommended and free models
GPT-6.1 Sol joins Cline’s recommended models
- Free list adds Solar Mini 4
- Free list drops DeepSeek V4.1 Flash
- Free list drops space-bunny-alpha
GPT 6.1 Sol joins Cline's recommended list. On the free list, Solar Mini 4 comes in, while DeepSeek V4.1 Flash and space-bunny-alpha drop off it.
Smaller fixes
Fixes for Kimi K3, tool approvals and reopened tasks
- Kimi K3 gets the closest reasoning level it supports
- Approve and Reject respond while the panel updates
- Reopened tasks drop leftover Focus Chain rows
Kimi K3 and other models that only accept certain reasoning levels stop rejecting requests. Cline picks the closest level they support. Approve and Reject on a tool approval work right away. Reopened tasks lose their leftover Focus Chain rows.
Release notes
Every provider’s new default is in the release notes
CHANGELOG.md in the cline/cline repository on GitHub
The full list of providers with a new default model is in the release notes. They live in the changelog in Cline's GitHub repository.










