SIGN IN SIGN UP

feat(provider/deepseek): complete vision content parts (#19423)

## Background

DeepSeek's Chat Completions schema supports image processing detail and
inline `file_data` content parts, but the provider only emitted
`image_url.url` or an uploaded `file_id`.

## Summary

- add typed `imageDetail` support for `low`, `high`, `original`, and
`auto`
- add an opt-in `fileData: true` representation for inline image bytes
and preserve filenames
- validate the documented JPEG/PNG/GIF/WebP formats and 8,192-character
HTTP URL limit
- keep the existing `image_url` behavior as the default

## Contributor Credit

Implemented by @gr2m with Codex.

## End-to-End Verification

- `pnpm --filter @ai-sdk/deepseek... build`
- `pnpm --filter @ai-sdk/deepseek test:node` — 76 tests passed
- `pnpm --filter @ai-sdk/deepseek test:edge` — 76 tests passed
- `pnpm --filter @ai-sdk/deepseek type-check`
- focused Ultracite check and `git diff --check`

## Checklist

- [x] Signed commit
- [x] Tests added/updated
- [x] Documentation updated
- [x] Patch changeset added
- [x] Self-reviewed

## Related Issues

Fixes #19384
G
Gregor Martynus committed
2814b2de18101495f9d664ef2f541e57933c0455
Parent: 3789213
Committed by GitHub <noreply@github.com> on 8/25/2026, 12:42:44 AM