Opencode + Qwen3.6 can not read image.

[lumixraku]

Description

Model qwen 3.6 can read image !
Claude code + Qwen 3.6 can read image, why it is not working with opencode?

Plugins

no plugin!

OpenCode version

1.15.10

Steps to reproduce

1. model change to Qwen3.6
2. paste an image

Screenshot and/or share link

No response

Operating System

macOS 26

Terminal

Warp

[github-actions]

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

• Qwen3.6-plus model throws "System message must be at the beginning" error with image input in OpenCode 1.3.13; Qwen3.5-plus works normally #20813: Qwen3.6-plus model throws error with image input (same model, same symptom)
• Unable to read images for some models #5359: Unable to read images for some models (general image reading issue across models)
• Read tool cannot pass image data to vision-capable models #15728: Read tool cannot pass image data to vision-capable models (includes Qwen models)
• opencode cannot read images anymore. #25832: opencode cannot read images anymore (general image input regression)

[samuel-pro-ai]

confirming this is a broader qwen family issue, not just qwen 3.6.

tested june 12 2026, opencode v1.17.3, windows + cursor terminal:

qwen3.6-plus on opencode-go: BROKEN

• direct chat attachment: "Cannot read clipboard (this model does not support image input)"
• read tool (file path): returns "Image read successfully" but model cant see the image

qwen3.7-plus on opencode-go: also BROKEN

• same error on both paths

for comparison:

• qwen3.7-max: couldnt even connect, server error
• mimo v2.5, m3, kimi k2.6 on the same provider: all work fine

so the qwen models specifically seem to be missing the MediaUnderstandingProvider registration in the go runtime. the underlying qwen models support images (zen prefix works on the same names) but the go provider drops them.

fwiw i tested with simple images (screenshots, photos) - nothing exotic. ~1MB jpgs and pngs. nothing weird about the input.

the user-level modalities config (per #9897) does not fix this. confirmed on m2.7 first, then confirmed again on qwen3.7-plus.

[xkaaaaaa]

不是模型的问题，我在 claude code 用同样的api 就是可以读取图像的

[michaeltg17]

This fix works #33695 (comment)