Read tool fails to process images with direct Anthropic provider (Claude Opus 4) on v1.2.15

[disolaterX]

Description

The Read tool fails to read image files with the error "Could not process image" when using Claude Opus 4 via the direct Anthropic provider. The model supports image input, but the Read tool's modality check appears to reject it.

Environment

• OpenCode version: 1.2.15
• Provider: Anthropic (direct, no proxy)
• Model: anthropic/claude-opus-4-6
• OS: macOS
• Plugin: oh-my-opencode@latest

Steps to Reproduce

1. Configure OpenCode with direct Anthropic provider and Claude Opus 4
2. Have any image file (PNG, JPG, etc.) in the working directory
3. Ask the model to read the image file using the Read tool
4. Read tool returns an error about not being able to process the image

Expected Behavior

The Read tool should detect that Claude Opus 4 supports image input and return the base64-encoded image as an attachment.

Actual Behavior

Read tool rejects the image with a "could not process image" error.

Related Issues

• Reading images #2657 (original image reading issue, fixed by PR feat: allow read tool to handle images #3052)
• [v0.14.7] Read tool incorrectly detects model as non-image-capable #3084 (model capability detection bug, closed)
• opus not able to read images with read tool #5056 (Opus not reading images, closed)
• Read tool fails to read image files even with vision-capable models (paste works) #11306 (Read tool fails for vision-capable models, closed)

This appears to be a regression or incomplete fix in the modality detection logic for direct Anthropic provider models.

[github-actions]

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

• model "claude-opus-4.6" not supported for vision #14289: Same symptom — Claude Opus 4.6 reported as not supporting vision/image input in OpenCode, with an active bug label and ongoing discussion
• Unable to read images for some models #5359: "Unable to read images for some models" — broader issue of the Read tool failing for vision-capable models
• Z.AI GLM-4.7 cannot read images from Read tool (model supports vision) #13310: "Z.AI GLM-4.7 cannot read images from Read tool (model supports vision)" — same Read tool + modality detection pattern for a different provider

If your issue is specifically about the Read tool path rather than drag-and-drop, it may still be distinct from #14289 and worth tracking separately, but please review those first.