#deepseek#vision#screenshot#agents#v4-flash-vision
DeepSeek V4 Flash Vision Exp: Screenshot-Aware Coding Agents Explained
DeepSeek's new V4 Flash Vision Exp model understands screenshots, images, and visual prompts. Learn how to send images via the API, how vision token billing works, and how it compares to the standard V4 Flash pricing.
#Table of Contents
DeepSeek just shipped its first widely available vision-capable model,
It's an experimental model that accepts images alongside text. You can ask it to:
All use the standard OpenAI-compatible format where
Images sent to
Here's how the pricing compares, per 1M tokens (peak/off-peak):
Key takeaway: The vision-exp model uses the same pricing as standard V4 Flash — there's no premium for image input. Payment is charged only on your consumption (top-up or granted balance), and peak hours are Mon–Fri 01:00–04:00 and 06:00–10:00 UTC.
deepseek-v4-flash-vision-exp, and it's a big deal for anyone building screenshot-aware coding agents or browser agents. It reads images and screenshots directly through the standard Chat Completions API — no special vision endpoint needed.
What Is deepseek-v4-flash-vision-exp?
- Describe pictures and analyze charts
- Read text from screenshots — perfect for UI automation, error messages, and visual review
- Understand visual prompts in agent workflows
How to Send an Image (3 Ways)
content is an array of blocks.
1. Base64-encoded image (inline)
The simplest option for local files. The encoded data counts toward the 48 MiB request body limit.import base64
from openai import OpenAI
client = OpenAI(api_key="<DeepSeek API Key>", base_url="https://api.deepseek.com")
with open("image.jpg", "rb") as f:
b64 = base64.b64encode(f.read()).decode("utf-8")
response = client.chat.completions.create(
model="deepseek-v4-flash-vision-exp",
messages=[{
"role": "user",
"content": [
{"type": "text", "text": "What is in this image?"},
{"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64}"}},
],
}],
)
print(response.choices[0].message.content)2. External image URL
Pass a publicly accessiblehttp(s) link. The URL must be ≤ 8192 chars, the image ≤ 32 MiB, and the download must complete within 60 seconds. For longer links, use base64 or the Files API.
3. Reference a file via the Files API
Upload an image once with the Files API, then reference it in your request. Good for large images or repeated use.Vision Token Billing
deepseek-v4-flash-vision-exp are converted into tokens based on their dimensions and billed as input tokens together with your text tokens. So the cost scales with image size.
Pricing: Vision Exp vs Standard DeepSeek
| Token Type | Model | Off-Peak | Peak |
|---|---|---|---|
| Input (cache hit) | V4 Flash Vision Exp | $0.007 | $0.014 |
| Input (cache miss) | V4 Flash Vision Exp | $0.22 | $0.44 |
| Output | V4 Flash Vision Exp | $0.66 | $1.32 |
| Input (cache miss) | V4 Flash (standard) | $0.22 | $0.44 |
| Output | V4 Flash (standard) | $0.66 | $1.32 |
Why This Matters for Agent Builders
- Screenshot-aware agents can now read UI screenshots directly, enabling visual regression and browser automation.
- Coding agents can interpret error screenshots, diagrams, and mockups without a separate OCR pipeline.
- It's an early test, not just a benchmark announcement — worth hands-on experimentation.
Recap
| Feature | Detail |
|---|---|
| Model | deepseek-v4-flash-vision-exp |
| Input formats | JPEG, PNG, GIF, WebP |
| Send image | Base64, URL, or Files API |
| Billing | Converted to tokens, billed as input |
| Pricing | Same as V4 Flash (no premium) |
| Use cases | Screenshot-aware agents, UI automation |
References
Share this research
<IE/>
Written by Ieproject
Tech hobbyist and explorer passionate about DeepSeek V4 Flash, Antigravity AI, OpenCode, Hermes Agent, and modern developer tools.
More Research & Guides
26 days agoIeproject