Does DeepSeek have a Files API?

Yes — since August 21, 2026. DeepSeek launched a Files API alongside the deepseek-v4-flash-vision-exp vision model. Upload a file with purpose=user_data, receive a file_id (e.g. file-api-xxxx), and reference it in chat messages with a file content block. Files are tied to your API key and usable from both the OpenAI-compatible and Anthropic-compatible APIs.

What are the DeepSeek Files API limits?

Per the official guide: a single file may be at most 64 MiB and the upload must finish within 10 minutes; max storage per user is 25 GiB; max 10,000 stored files per user; retention is 1 hour to 30 days, or permanent if you omit expires_after. Supported formats are JPEG, PNG, GIF, and WebP, detected by actual file content rather than extension.

How do I reference an uploaded file in a chat request?

Include a content block of type "file" with the file_id returned by the upload, in the message content array alongside text blocks — used with the deepseek-v4-flash-vision-exp model. From the Anthropic-compatible /messages endpoint you also need the anthropic-beta: files-api-2025-04-14 header. Alternatively, a file block can carry base64 data inline via file_data instead of file_id; the two fields are mutually exclusive.

Why use the Files API instead of base64 inline images?

Two reasons: size and repetition. Inline base64 images are capped at 32 MiB per image, while a file_id image may be up to 64 MiB in a request; and agent loops that resend the same screenshots every turn can reference a file_id instead of re-uploading megabytes each request — smaller payloads, faster requests, cleaner caches.

Guides / DeepSeek Files API

Guide · verified 2026-08-21 against api-docs.deepseek.com

DeepSeek Files API (2026): Limits, Uploads, and file_id in Practice

DeepSeek quietly shipped its first file endpoint on August 21, 2026, alongside the vision model — which makes a lot of older guides obsolete (several still claim DeepSeek's API has "no file upload route"). This guide has the verified limits, upload and reference examples in curl and Python, and when to use files instead of base64.

Direct answer

The DeepSeek Files API uploads an image once and returns a file_id you reference in chat. Upload with purpose="user_data"; a file may be up to 64 MiB (10-minute upload window); storage is 25 GiB / 10,000 files per user; retention 1 hour–30 days or permanent without expires_after. Reference files from deepseek-v4-flash-vision-exp chats via a {"type":"file","file_id":...} block — file_id images can be 64 MiB in a request versus 32 MiB inline. From the Anthropic-compatible endpoint, add the anthropic-beta: files-api-2025-04-14 header.

Limits at a glance

Limit	Value (official docs, 2026-08-21)
Formats	JPEG, PNG, GIF, WebP — detected by file content, not extension
Per file	Max 64 MiB; upload must complete within 10 minutes
Per user	25 GiB storage; 10,000 stored files
Retention	1 hour to 30 days, or permanent (omit expires_after)
Purpose	Must be `user_data` (only supported purpose, also for list filtering)
In-request size	file_id image up to 64 MiB per request; inline base64/URL images capped at 32 MiB
API compatibility	OpenAI-compatible and Anthropic-compatible; Anthropic /messages needs the files-api beta header

Source: official Files API guide, read 2026-08-21.

Upload once, reference forever (well, up to 30 days)

Upload a screenshot and get a file_id:

curl https://api.deepseek.com/files \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -F "purpose=user_data" \
  -F "file=@dashboard.png"

Then reference it in a vision-exp chat — no base64 blob, no re-upload on subsequent turns:

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -d '{
    "model": "deepseek-v4-flash-vision-exp",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Which metric regressed in this dashboard?"},
        {"type": "file", "file_id": "file-api-xxxx"}
      ]
    }]
  }'

Python with the OpenAI SDK — upload, then chat:

from openai import OpenAI

client = OpenAI(api_key="YOUR_KEY", base_url="https://api.deepseek.com")

f = client.files.create(file=open("dashboard.png", "rb"), purpose="user_data")

resp = client.chat.completions.create(
    model="deepseek-v4-flash-vision-exp",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "Which metric regressed?"},
            {"type": "file", "file_id": f.id},
        ],
    }],
)
print(resp.choices[0].message.content)

Inline alternative: a file block can carry file_data (base64) instead of file_id — mutually exclusive fields; filename may only accompany file_data.
Anthropic-style stack: same file blocks work on /messages with the anthropic-beta: files-api-2025-04-14 header.

When to use files vs base64 vs URL

Method	Max size	Best for
base64 inline	32 MiB	One-off images generated in-process; simplest to code
external URL	32 MiB, 8,192-char URL, 60s fetch	Publicly hosted images; no upload step at all
Files API	64 MiB	Agent screenshot loops, repeated reference images, anything reused across turns

FAQ

I read DeepSeek has no file endpoint — is this new?

Correct until Aug 21, 2026 — older guides describing "no /v1/files route, extract text yourself" were accurate for their time and are now outdated. The Files API shipped with the vision launch.

Can I upload PDFs or documents?

No — the Files API currently accepts image formats only (JPEG/PNG/GIF/WebP). For PDFs, render pages to images caller-side, then upload or inline them.