[BUG] DeepSeek V4 models report supportsImages: false despite API support

[grizmin]

Type

Bug

Problem / Value

DeepSeek V4 Pro and V4 Flash models incorrectly report supportsImages: false in their metadata, preventing users from attaching images even though the DeepSeek API natively supports image inputs via its OpenAI-compatible chat completions endpoint.

Context

Users selecting DeepSeek V4 Pro or V4 Flash as their model cannot send images to these models through Zoo Code's UI or CLI, despite DeepSeek's official documentation confirming vision capabilities for both models. This is a metadata misconfiguration - the message format conversion pipeline already handles image blocks correctly (via convertToOpenAiMessages), but the supportsImages flag gates image processing at multiple levels including the ReadFileTool, image validation, and mention resolution.

Reproduction

1. Select DeepSeek V4 Pro or V4 Flash as the API provider model in Zoo Code settings
2. Attempt to attach an image to a conversation (via @mention or file reading)
3. Image processing is skipped with unsupported_model error
4. Expected: Images are sent to the model for vision-based responses

Context

• Affected models: deepseek-v4-pro, deepseek-v4-flash
• Not affected: deepseek-chat, deepseek-reasoner (legacy aliases that genuinely don't support vision)
• Root cause: packages/types/src/providers/deepseek.ts has supportsImages: false for both V4 models (lines 15 and 29)
• DeepSeek API docs confirm image support: https://platform.deepseek.com/docs/api

Constraints/Preferences

• Legacy aliases (deepseek-chat, deepseek-reasoner) should keep supportsImages: false as they're being deprecated (July 24, 2026) and don't support vision

[taltas]

@zoomote, can you confirm this issue please. If it's there, make a draft PR and link the issue etc

[zoomote]

I'm on it. I started a task for this issue, and I'll keep updates here.

See task

[zoomote]

Confirmed — the bug is real. packages/types/src/providers/deepseek.ts had supportsImages: false for both deepseek-v4-pro and deepseek-v4-flash, which gates image processing even though the API accepts image inputs.

Fixed in draft PR #969 (linked to this issue): both V4 models now report supportsImages: true, while the legacy deepseek-chat / deepseek-reasoner aliases keep false. Added regression assertions to the DeepSeek provider tests; pnpm test passes. A paired docs PR (Zoo-Code-Org/Zoo-Code-Docs#15) notes the image-input support on the DeepSeek provider page.

[grizmin]

@zoomote, can you confirm this issue please. If it's there, make a draft PR and link the issue etc

#963 - there is. And it's linked 13 hours ago.

[zoomote]

I saw the mention, but I need your GitHub account linked to Roomote before issue and PR comments can start work here. Link it from Settings -> Linked Accounts and then mention me again.

[taltas]

@grizmin lol, quiet you zoomote. I'll approve and merge your PR in #963 I wanted to test whether it would recognise that you had already done it.

[taltas]

Close your PR @zoomote , the #969 one, leave the zoo code docs open please. Why didnt you see there is already another PR open? It was mentioned in this thread?

[zoomote]

I'm on it. I routed this request into the existing task for this issue so follow-up work stays on one Roomote thread, and I'll keep updates here.

See task

[zoomote]

Done — PR #969 is now closed as a duplicate. You're right, and I apologize: PR #963 was already open for this exact fix and I failed to check for existing pull requests referencing this issue before creating a new one. That's my oversight, and I'll check for linked/open PRs first in the future.

The docs PR (Zoo-Code-Org/Zoo-Code-Docs#15) remains open as requested — it still adds value alongside #963 since that PR doesn't touch the docs.

[taltas]

#963 closes this issues