Cline SDK Update Adds OpenAI Compatibility and Model Enhancements

What happened

The latest Cline SDK update now supports OpenAI-compatible requests for GPT-6 models, which use the parameter 'max_completion_tokens' instead of 'max_tokens'. This change aligns with OpenAI's newer model naming conventions.

Azure's 'gpt-35-turbo' model continues to use the 'max_tokens' parameter, indicating a distinction in how different platforms handle token limits. The hub daemon now records the client that initiated it and publishes this information as OTel resource attributes, offering more visibility into system interactions.

Why it matters

These updates improve compatibility with various AI infrastructure platforms, enabling developers to integrate models more seamlessly. The enhanced telemetry system now includes 'traces_exporter', which improves data collection and monitoring for distributed applications.

Key details

The model catalog has been expanded to include 7,041 models, up from 6,952. New entries include the CommandCode and Reflection providers, which offer additional tools for developers. The Cline recommended list has reverted to using the Claude Sonnet 5.5 model as the default for the 'cline' provider.

Default models for several platforms have been updated, including the Gemini 4 Argon model for 302.AI, the Muse Spark Open model for Eden AI and Prime Inference, and the Reflection Beam 1.0 model for CoreWeave. The OpenCode Go model now uses a pinned base URL, reflecting changes in Models.dev's provider-level URL structure.