Cline v4.1.22 Adds New AI Providers, Fixes Anthropic Failover and Bedrock Routing Bugs

Cline, the open-source AI coding agent for VS Code, shipped version 4.1.22 on September 30, adding two new inference providers and refreshing default models across a wide swath of third-party gateways, according to the GitHub release notes.

New providers

The release adds support for two new providers: Bee (by HEOSSI) and Pareto Inference, the latter accessible via a PARETO_API_KEY environment variable.

Default model changes

Cline refreshed its model catalog across many gateway integrations. GPT-6.1 Sol is now the default model for OpenAI, OpenRouter, GitHub Copilot, Cortecs, Eden AI, Kilo Gateway, both LLM Gateway providers, NanoGPT, OpenCode Zen, and Requesty. Other gateway defaults shifted as well: Vercel AI Gateway now defaults to Ling 3.1 Flash, Tempr Gateway to MiMo V2.6 Flash, CrossModel and Ofox to Claude Sonnet 5.5, Pioneer to GLiNER 2.5 Decide, and Scaleway to Qwen 3.8 27B. Cline's recommended model list now includes Claude Sonnet 5.5 and Claude Opus 5.5. Users who have not pinned a specific model on any of these providers will see a different default model going forward.

Reliability changes

Direct requests to Anthropic now use Anthropic's server-side refusal fallback. On OpenRouter and Cline's own routing, Anthropic models can fail over to an alternate upstream provider rather than failing the request outright. When a response is blocked by a content filter, Cline now surfaces that explicitly and suggests rephrasing, instead of showing a generic "Model returned empty response" error.

Bug fixes

The release fixes a cluster of issues affecting task recovery and messaging: users can now type and send a new message after an API error instead of being limited to Retry; commands like /compact, /smol, or /newtask typed during recovery stay in the composer rather than being sent to the model as plain text; and a failed sign-in during a new task no longer silently drops a revised prompt. "Reset Code" now correctly disables when a message has no checkpoint and no longer stays stuck disabled on slow disks. Checkpoint setting changes now apply to the active task without losing queued follow-up messages, and canceling a task immediately after sending a message now reliably stops it — previously a cancel that landed during turn setup was ignored. Reasoning tokens are no longer double-counted in token usage displays, though reported cost is unchanged.

On Amazon Bedrock, GPT-6 and GPT-5.6 now route through inference profiles, with India regions (ap-south-1, ap-south-2) resolving to the correct "in." profile prefix. OpenAI models behind inference profiles now receive the reasoning-effort setting they support. Requests that Bedrock previously rejected for Nova 2 Lite with high reasoning, application inference-profile ARNs, and Nova Micro are now fixed. A legacy AWS profile setting saved without the "use profile" flag now correctly carries over as profile authentication instead of being dropped. Gateway models on providers that mix API endpoints now retain their own protocol instead of falling back to the provider-wide default.

Full changelog: v4.1.21...v4.1.22

What this means

This is a maintenance release for Cline's routing and provider layer, not a new model. The notable signal is how much of Cline's value now depends on third-party model availability and defaults — a single release note lists default-model swaps across more than a dozen gateways, meaning users who rely on unpinned defaults can see behavior change without any action on their part. The Anthropic failover and content-filter messaging changes address a common pain point in production coding agents: silent failures that look like model bugs but are actually infrastructure issues. Teams running Cline in CI or on Bedrock should note the inference-profile and reasoning-effort fixes, which resolve requests that were previously rejected outright.