Cline v4.1.21 Fixes Local Model Timeouts, Expands Catalog to 6,386 Models Across 209 Providers
Cline's v4.1.21 update adds a new OpenAI-compatible provider called ai&, refreshes its model catalog to 6,386 models across 209 providers, and fixes a bug that ended tasks prematurely when local models hit output-token limits. Eleven unpinned providers, including GitHub Copilot and Vertex, now default to Claude Opus 5.5.
What shipped
Cline, the open-source AI coding agent for VS Code, released version 4.1.21 on September 24. The update focuses on provider catalog maintenance, a security patch, and several reliability fixes for local model workflows.
New provider and catalog refresh
The release adds support for ai&, an OpenAI-compatible endpoint serving open-weight models from Japan. Cline's model catalog now lists 6,386 models across 209 providers, according to the release notes.
As part of this refresh, the resolved default model changes for 19 providers that don't pin a specific model version. Eleven of those, including GitHub Copilot and Vertex, now default to Claude Opus 5.5. Developers relying on unpinned defaults through these providers will see a different model behind their requests after upgrading.
Security fix
Cline raised its minimum js-yaml dependency to version 4.3.2, addressing a security fix in the parser used to read rule and skill frontmatter files.
Local model reliability fixes
The most substantive fix targets local model backends: llama.cpp, Ollama, and LM Studio. These servers cap generation length based on whatever context space remains, independent of the output-token budget a user sets. Previously, long replies that hit this limit would end the task outright. Now, Cline compacts the conversation and retries once; if compaction doesn't resolve the issue, the existing concise-retry recovery mechanism still runs, and any partial answer generated is preserved rather than discarded.
Other fixes
- Reopening a task that previously failed now correctly displays the error with a retry option, rather than presenting the failed task as completed.
- Commands that produce no output no longer display a raw JSON artifact (e.g.,
[{"query":"git add -A","result":"","success":true}]) in place of actual output. - On Windows, @-mention file search now returns correct filenames; nested files (such as a subfolder's README) previously rendered as
/README.md, and duplicate entries for the same open file have been eliminated. - Canceling a request while it's waiting to retry after an empty model response now takes effect immediately instead of waiting for the backoff period to finish.
The full changelog is available on GitHub comparing v4.1.20 to v4.1.21.
What this means
This is a maintenance release, not a new model or major feature launch. The most consequential item for teams running Cline in production is the default-model shift to Claude Opus 5.5 for 11 unpinned providers — anyone who hasn't explicitly pinned a model version through GitHub Copilot, Vertex, or the other affected providers will see behavior and cost changes without any local configuration change on their end. The local-model output-limit fix addresses a real pain point for developers running llama.cpp, Ollama, or LM Studio, where previously a single long response could silently kill an entire task instead of recovering gracefully. Together, these changes signal Cline's continued focus on multi-provider stability rather than expanding its own model or feature surface.
Related Articles
OpenAI Python SDK v3.56.0 adds gpt-live-1.1 to Live model choices
OpenAI's Python SDK v3.56.0, dated 2026-11-19, adds gpt-live-1.1 to the Live model choices. The release lists this as its only feature.
DeepSeek removes separate fees for model-triggered search and web fetch from Nov 18
DeepSeek removed separate charges for server-side search and web fetch triggered during model calls, effective at noon on November 18, 2026, Beijing time. Token usage is still billed, and the standalone Search and Fetch APIs remain at 5 yuan per 1,000 successful calls.
iOS 27.3 adds daily usage limits to advanced Siri AI requests; basic requests stay unlimited
Apple's iOS 27.3 introduces usage limits for advanced Siri AI tasks, and basic requests are not limited. A leaker has published the criteria Siri AI uses to classify a request as advanced, while Apple has shared only two examples.
Gemini Skills now support sharing and live Google Drive file references, rolling out to all Workspace and personal accou
Google has added sharing and file attachments to Skills, the custom-instruction feature in the Gemini app. Files from Google Drive and Gemini Notebook update in real time, and each skill accepts up to 100MB of uploaded files. The rollout covers all Google Workspace and personal Google account users.
Comments
Loading...