OpenCode Go fails with GLM-5.3 and GLM-5.3-flash when used through Claude Code. #51120

Activity

1. ljluestc

   added a commit that references this issue on Oct 4, 2026

   docs(go): note that each model only works on its listed endpoint

   ...

   This issue will close once commit 647c034 is merged into the 'dev' branch.

   647c034
2. poppy-mogmachine commented on Oct 15, 2026

   

   poppy-mogmachine

   on Oct 15, 2026

   More actions

   The {"model":"glm-5.3"} in your report is the clue: on Go, GLM is only reachable through /v1/chat/completions, and the /v1/messages route returns 503 for it (#50502 lists the same asymmetry for MiMo, kimi-k2.x and others). Claude Code only speaks /v1/messages, so it won't work until the server side is fixed.

   Until then, you can put a small local LiteLLM proxy in front. It accepts Anthropic Messages from Claude Code and sends Chat Completions upstream:

   # litellm.yaml
   model_list:
     - model_name: glm-5.3
       litellm_params:
         model: openai/glm-5.3
         api_base: https://opencode.ai/zen/go/v1
         api_key: os.environ/OPENCODE_GO_KEY
     - model_name: glm-5.3-flash
       litellm_params:
         model: openai/glm-5.3-flash
         api_base: https://opencode.ai/zen/go/v1
         api_key: os.environ/OPENCODE_GO_KEY
   litellm_settings:
     use_chat_completions_url_for_anthropic_messages: true   # the important line
   general_settings:
     master_key: os.environ/LITELLM_MASTER_KEY

   litellm --config litellm.yaml --port 4000
   export ANTHROPIC_BASE_URL=http://localhost:4000
   export ANTHROPIC_AUTH_TOKEN=$LITELLM_MASTER_KEY
   export ANTHROPIC_DEFAULT_OPUS_MODEL=glm-5.3
   export ANTHROPIC_DEFAULT_SONNET_MODEL=glm-5.3-flash

   If you leave out use_chat_completions_url_for_anthropic_messages, LiteLLM passes /v1/messages straight through and you get the same error back. I tested this on LiteLLM 1.105.1 against a different chat-completions-only GLM-5.3 upstream, not Go itself (I don't have a Go key). Both tool_use and streaming came back as proper Anthropic events.
3. neriousy commented on Oct 20, 2026

   

   neriousy

   on Oct 20, 2026

   Member

   More actions

   Thanks for the report. We're tracking this same Go /v1/messages failure for GLM models in #50502, so I'm closing this as a duplicate. Please follow that issue for updates.
4. neriousy

   closed this as a duplicateof go: /v1/messages returns 503 "Endpoint is unavailable" for mimo/glm and other models while /v1/chat/completions works #50502 on Oct 20, 2026