Cline v4.1.25 adds cloud sandbox sessions and makes Gemini 4 Argon the default model for its own provider

Cline v4.1.25 adds Cline Cloud sessions, a feature that runs tasks in a hosted sandbox on a GitHub repo and branch the user selects. The release also changes the default model for the Cline provider and for more than 25 other providers. According to the release notes, cloud sessions are rolling out gradually and are enabled in Settings.

Cline Cloud sessions

Cloud tasks execute in a hosted sandbox instead of the local editor. Per the release notes:

• Tasks appear in Recent and History with their status.
• Cline notifies the user when a task finishes.
• Tasks keep running after the editor is closed.

The notes do not disclose pricing, usage limits, sandbox specifications, or which plans have access.

Default model changes

With no model chosen, the Cline provider now defaults to Gemini 4 Argon, which becomes the first entry in Cline's recommended list. OpenRouter's default remains Claude Sonnet 5.5. The release notes give no pricing, context window, or benchmark data for any of the models named.

Defaults changed for these providers:

• Gemini 4 Argon: Cortecs, CrossModel, Eden AI, Google Gemini, GitHub Copilot, Google Vertex AI, LLM Gateway, Ofox, OpenCode Go, and Prime Inference.
• Muse Spark Open: Fireworks AI, Nebius Token Factory, and Requesty.
• Nemotron 3.5 Super VL: CoreWeave, DigitalOcean, NanoGPT, and Nvidia.
• Claude Haiku 5.5: Aixy and Neon.
• Changed, new default not specified in the notes: Agnes AI, Ambient, IteraCompute, Poe, Scaleway, and SenseNova.

Users on any of these providers who have not pinned a model will get a different default after updating.

Catalog and list changes

• The model catalog was refreshed, adding the Boundless and Standard Compute providers.
• SambaNova now pulls from the catalog, with MiniMax M3 as its default.
• Cline's recommended list adds Gemini 4 Argon and drops Grok 4.7.
• Cline's free list replaces Step 5 Preview with Nemotron 3.5 Super VL.
• OpenCode Go no longer lists MiniMax M2.7.

Fixes

• Local model timeouts: Long requests to LM Studio and Ollama no longer fail with terminated: BodyTimeoutError after 5 minutes of silence while the server is still processing a large prompt.
• OpenAI Compatible base URLs: URLs are now used exactly as entered. Previously, a host-only URL such as https://abc.com had /v1 appended silently. Host-only URLs saved by earlier versions get /v1 added once, so existing setups continue to work.
• Attachments: Images and files attached to an answer to one of Cline's questions now reach the model instead of being dropped.
• Stream retries: If a provider error interrupts a response mid-stream, Cline retries up to 3 times before failing the turn.
• JetBrains UI: The Run task card now stays above the Auto-approve bar.
• Feature settings: Auto Compact and its strategy setting sit together again.

What this means

Cloud sessions move Cline from a purely local editor agent toward hosted, asynchronous execution, a model already used by other coding-agent products. The practical effect is that long tasks no longer depend on an open editor. Until pricing, isolation details, and rollout scope are published, teams should treat it as a preview.

The default-model changes matter more for day-to-day behavior than they appear to. Many Cline users never pin a model, so a default swap across 25+ providers changes output quality, speed, and cost for them without any action on their part. Anyone with reproducibility or budget requirements should pin a model explicitly before upgrading.

The base URL change is a quiet breaking-behavior fix. Self-hosted gateways that do not serve under /v1 previously failed because of the silent rewrite. The one-time migration for saved host-only URLs reduces the risk for existing configurations.

The local model timeout fix removes a common failure for users running large prompts through Ollama or LM Studio on modest hardware.