ttachment: true in custom provider model config has no effect - image upload blocked with 'does not support image input' #33542

Activity

1. github-actions

   assigned

   jlongster

   on Jun 23, 2026
2. github-actions commented on Jun 23, 2026

   

   github-actionsBot

   on Jun 23, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue might be a duplicate of existing issues. Please check:

   • Bug: qwen3.6-plus model in alibaba-coding-plan-cn provider has attachment: false, preventing image input #28690: Same symptom — attachment: true config being ignored, image input blocked at the frontend level (workaround described there matches your config)
   • Custom OpenAI-compatible providers: image file attachments do not reach vision-capable models correctly #20802: Very similar — custom provider image attachments not reaching vision-capable models, same error message ("this model does not support image input")
3. cipherfoxie commented on Jul 6, 2026

   

   cipherfoxie

   on Jul 6, 2026

   More actions

   Update: Confirmed on local Qwen3.6-35B via vLLM — attachment: true still blocks images at frontend.

   Environment:

   • OpenCode: 1.17.13 (Go binary, aarch64)
   • Model: Qwen3.6-35B-A3B via vLLM localhost:30001
   • Config: "attachment": true set on model + "attachment.image" config at root level
   • OS: Linux ARM64 (NVIDIA DGX Spark, 121 GB Unified Memory)

   Test results:

   1. "attachment": true in model config → still blocked by frontend with "does not support image input"
   2. "attachment.image" root config (auto_resize, max_width/height, max_base64_bytes) → applied but irrelevant since image never reaches API
   3. Direct vLLM API call with base64 image → works perfectly, Qwen describes image correctly
   4. webfetch via URL → works but I cannot "see" images (only metadata), not real vision

   Workaround that works:

   python3 - <<'PY'
   import json, urllib.request
   req = {
     "model": "qwen3.6-35b",
     "max_tokens": 60, "temperature": 0,
     "messages": [{
       "role": "user",
       "content": [
         {"type": "text", "text": "Beschreib das Bild."},
         {"type": "image_url", "image_url": {"url": "<image_url>"}}
       ]
     }]
   }
   r = urllib.request.urlopen(urllib.request.Request(
     "http://localhost:30001/v1/chat/completions",
     data=json.dumps(req).encode(),
     headers={"Content-Type": "application/json"}
   ), timeout=120)
   d = json.load(r)
   print(d["choices"][0]["message"]["content"])
   PY

   Root cause hypothesis: The frontend checks model vision capability against the provider's models metadata (likely via /v1/models endpoint or hardcoded list), and attachment: true in opencode.json is not consulted. The check happens before the request is built, so the image is dropped client-side.

   Suggestion: The frontend should check both:

   1. Provider/model metadata for vision support, AND
   2. "attachment": true in the model config as an override

   Until fixed, the workaround is calling the vLLM API directly with base64 images. This is functional but breaks the opencode workflow — images cannot be part of the conversation context.
4. github-actions commented on Sep 5, 2026

   

   github-actionsBot

   on Sep 5, 2026 – with GitHub Actions

   Contributor

   More actions

   To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.
5. github-actions

   closed this as not plannedon Sep 5, 2026