Claude Desktop raises the wall: Gateway update blocks non-Claude models, community gameplay becomes useless in two weeks

robot
Abstract generation in progress

According to Beating monitoring, Claude Desktop pushed version 1.6259.1 on May 6, adding model ID whitelist filtering to the third-party inference (3P) Gateway mode. Previously, the community used Gateway interfaces to connect to gateways like OpenRouter and run non-Claude models such as GPT, Grok, and DeepSeek, but after the update, all of those methods failed.

After the update, the model discovery interface in Gateway mode checks every model ID returned by /v1/models, keeping only entries that start with claude or anthropic. Ollama users reported that the gateway returned 39 models, but after filtering in Claude Desktop, 0 available models were shown. Manually configuring non-Anthropic models in the inferenceModels configuration also caused errors: “configured model is not an Anthropic model. Gateway deployments require an Anthropic model from the provider catalog.” Some users set up a local proxy and mapped the IDs of models like Kimi K2.6 into the claude-sonnet-4-6 format, restoring normal operation—proving that the restriction is at the model ID filtering layer and does not involve protocol or capability detection.

On the same day, the Claude Code CLI 2.1.129 release also tightened Gateway integration. Model discovery changed from the default automatic behavior in version 2.1.126 to opt-in, and users must manually set the environment variable CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1 to query the gateway’s model list.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin