Claude Desktop can now easily run Qwen, DeepSeek and Kimi models — after Ollama’s first effort stalled
The New Stack Paul Sawers
Claude Desktop can now route requests to Ollama models like Qwen, DeepSeek and Kimi. Ollama’s second try works because it slipped in a local proxy after the first one broke.
Based on reporting by The New Stack, Paul Sawers — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Ollama has brought back Claude Desktop support, and this time the link sticks. The open-weight model runner now lets Anthropic’s desktop app talk to models served through Ollama, including ones running on a user’s own machine. That means Claude users can stay inside the same interface while Ollama acts as the gateway to models like Qwen, DeepSeek, Kimi and GLM.
This is not Ollama’s first pass at the idea. In January, it added support for Anthropic’s Messages API so Claude Code could point at Ollama and use local or cloud-hosted models instead of Claude. Then in April, Anthropic quietly opened the door a little wider in Claude Desktop with a developer setting for third-party inference gateways. Ollama moved fast with a launch command that wired the two together, but the setup only lasted a few days before a Claude Desktop update started rejecting non-Anthropic model IDs. Ollama pulled the feature and said the integration had been reduced to Anthropic models only.
The new version, released in v0.33.0 on Aug. 21, takes a different route. Ollama says it now uses a dedicated local proxy for Claude Desktop, which gets around the mismatch that broke the first attempt. The setup is handled inside Ollama itself: open the app, choose Claude, turn on the integration, and it configures Claude Desktop’s third-party gateway automatically. Turn it off, and the previous Claude setup comes back.
Ollama has also tucked the toggle into its Mac menu, where users can switch “Use Ollama models” on or off. Once enabled, models available through Ollama, whether local or on Ollama Cloud, show up in Claude Desktop’s model picker. Users can even map the names Claude shows to different underlying models. In Ollama’s example, “Opus 5” routes to Kimi K3 and “Sonnet 5” maps to DeepSeek V4 Pro.
The point, Ollama says, is choice: cost, speed, portability, and the option to use models fine-tuned on a developer’s own data. Claude Desktop also keeps its own Auto mode, and Ollama’s integration leaves that alone. For now the feature is Mac-only, though Ollama has hinted on X that Windows support may be coming. The new push lands less than two months after Ollama raised $65 million.
My take — AI-written commentary, not fact-checked reporting
This is the kind of feature that quietly matters more than the demo clips do. The real fight in AI is not just model quality, it’s whether developers can swap models without rebuilding their entire workflow every time a vendor blinks. Claude Desktop acting as a front end for open models is a neat reminder that closed products get a lot more interesting when they’re forced to play nice.
Read more about this at: The New Stack
