DeepSeek V4 Flash Vision Exp

deepseek-v4-flash-vision-exp

Get API keyAPI reference

Use the experimental deepseek-v4-flash-vision-exp model to perform mixed image and text tasks, such as screenshot inspection, document extraction, chart analysis, and visual agent.

Context window1.0M

ProviderDeepSeek

Released2026/08/24

Playground

Pricing

The same model is available through multiple service channels — choose based on latency, reliability and cost.

Prices in $ / 1M tokens

To pick a channel, add a provider field to the request body, for example "provider": { "channel": "direct" }. Valid values are direct / stable / economical; omit it to use the default channel.

Direct

Direct upstream connection — best when you need native behavior and the full context window.

Input	Output	Cache read
0.44/M	1.32/M	0.01/M

Overview

Input

Text

Output

Text

DeepSeek V4 Flash Vision Exp API: Multimodal Experimental Model That Gives Flash Eyes

DeepSeek V4 Flash Vision Exp is an experimental multimodal model launched by DeepSeek on August 21, 2026, adding image understanding capabilities on top of V4 Flash. Text capabilities — including agent, reasoning, and world knowledge — are fully on par with V4 Flash, meaning adding vision does not sacrifice existing text performance.

The most notable change is its leap in multimodal agent benchmarks: overall performance approaches Claude Opus 4.8, surpassing it in 3 of 11 benchmarks. ApexBench (Pass@1) jumps from 26.2 for the text-only V4 Flash to 36.5, and Terminal Bench 2.1 reaches 83.9 (V4 Flash: 82.7, Opus 4.8: 85.0) — while it remains the same efficient sparse MoE with 13B activated parameters.

SeaWhale AI provides DeepSeek V4 Flash Vision Exp through an OpenAI-compatible interface, supporting image input, tool calling, and streaming output.

Get API Key · Model ID: deepseek-v4-flash-vision-exp

---

Why Choose DeepSeek V4 Flash Vision Exp

• Multimodal agent approaches Opus 4.8 — surpasses it in 3 of 11 benchmarks, yet priced at the Flash tier
• Zero loss in text capabilities — agent, reasoning, and world knowledge on par with V4 Flash
• Up to 384 tokens per image — image billing is capped regardless of resolution, so costs are predictable
• 284B total parameters / 13B activated — the same efficient sparse MoE as Flash
• 1M token context — long mixed text-and-image documents fit in a single pass
• Up to 600 images per request — no need to split large-scale image tasks across requests

---

Core Capabilities

01 Multimodal Agent

This is the core of this upgrade. With vision added, the model can directly "see" screenshots, interfaces, and charts in agent workflows to decide its next action, and its performance on multimodal agent benchmarks is now close to top closed-source models.

• ApexBench (Pass@1) 36.5, more than 10 points above text-only Flash
• Terminal Bench 2.1 reaches 83.9, surpassing text-only Flash
• Automated operation workflows driven by screenshot understanding

02 Image Understanding and Information Extraction

Supports three core vision tasks — image description, text extraction from screenshots, and chart analysis — covering the main scenarios of everyday multimodal needs.

• Text extraction from screenshots and scanned documents
• Structured interpretation of charts and diagrams
• Comprehensive understanding of mixed text-and-image documents

03 Controllable Image Costs

Each image is billed for at most 384 tokens, regardless of resolution; the detail field can also be used to further reduce token consumption for images that don't require high detail. Images are priced at the V4 Flash tier.

• Per-image token cap of 384, cost ceiling
• detail field reduces consumption on demand
• Total cost of large-scale image processing can be precisely estimated

04 Flexible Image Input Methods

Supports four formats: JPEG, PNG, GIF, and WebP, detecting format by actual file content rather than extension. Images can be passed via embedded Base64, public URLs (up to 32 MiB), or the free Files API (up to 64 MiB) — with the Files API, a single upload can be reused across requests via file_id, avoiding repeated transfers.

• Three input methods: Base64 / URL / Files API
• Files API is free and supports cross-request reuse
• Up to 600 images per request, max long edge 8192 pixels (4096 for more than 15 images)

---

Best Use Cases

Scenario	Description
Multimodal agent	Automated workflows that operate interfaces by viewing screenshots and make decisions by reading charts
Document digitization	Batch text extraction and structuring from scanned documents and screenshots
Chart analysis	Interpretation of reports, diagrams, and technical drawings
Mixed text-and-image long documents	Multimodal document understanding within a 1M token window
Large-scale image pipelines	600 images per request + per-image cost cap, suitable for scaled processing
Low-cost vision prototyping	Validate multimodal product ideas at Flash-tier pricing

---

Differences Between DeepSeek V4 Flash Vision Exp and V4 Flash

Capability	V4 Flash Vision Exp	V4 Flash
Model ID	deepseek-v4-flash-vision-exp	deepseek-v4-flash
Input modality	Text + image	Text only
Text capabilities	On par with Flash	Baseline
ApexBench (Pass@1)	36.5	26.2
Terminal Bench 2.1	83.9	82.7
Total / activated parameters	284B / 13B	284B / 13B
Context window	1M tokens	1M tokens
Max output	384K token	384K token
Status	Experimental (Exp)	Stable

For specific billing, refer to the real-time price card at the top of the page.

---

FAQ

What is its relationship with V4 Flash? It is the multimodal experimental version of V4 Flash: the same 284B/13B sparse MoE base, text capabilities on par, with image input added. Think of it as "Flash that can see images."

How good is its multimodal performance? It approaches Claude Opus 4.8 overall on multimodal agent benchmarks, surpassing it in 3 of 11. Representative data: ApexBench (Pass@1) 36.5, Terminal Bench 2.1 83.9.

How is image billing calculated? Each image is converted to at most 384 tokens, regardless of resolution, priced at the V4 Flash tier. For images that don't require high detail, use the detail field to further save tokens.

What are the image input limits? Supports JPEG, PNG, GIF, WebP; up to 600 images per request; max long edge 8192 pixels (reduced to 4096 when more than 15 images); max file size 32 MiB via URL and 64 MiB via Files API. Images are normalized to approximately 800×800 pixels before processing.

What does the "Exp" suffix mean? It is an experimental version DeepSeek uses to validate the multimodal direction; the interface and behavior may be adjusted upon the stable release. In production, it is recommended to make the model ID configurable for a smooth future switch to the stable version.

Context and output limits? 1M token context and max output of 384K token, the same as V4 Flash.

---

Why Use DeepSeek V4 Flash Vision Exp API with SeaWhale AI

• No self-built cluster needed — callable without provisioning your own GPU resources
• OpenAI-compatible interface — integrate with existing multimodal code with almost zero changes
• Compare with closed-source models in the same account — benchmark directly against Claude and GPT vision capabilities
• Settlement in RMB — domestic models billed in RMB

API

API integration

Model IDUse this value as the model in inference requests

deepseek-v4-flash-vision-exp

API KeyBearer token used to authenticate inference requests

Manage API keys

Base URLOpenAI compatible · /chat/completions

https://api.seawhaleai.com/v2/

provider OptionalSelects a service channel; omit it and the system picks the default

"provider": { "channel": "direct" }

deepseek-v4-flash-vision-exp usage examples

SeaWhale AI is compatible with the OpenAI API protocol, so you can call it with the OpenAI SDK or plain HTTP requests. Streaming is enabled by default.

About the provider parameter (optional, a SeaWhale AI extension): most models are served over several channels that differ slightly in price and reliability. Add a provider field to the request body to pick one; omit it and the system selects the default channel — normal calls are unaffected.

provider is not part of the official OpenAI protocol — it is a SeaWhale AI extension that only takes effect on this platform. The OpenAI SDK allows custom fields like this to pass through; see the examples below.

Value	Channel	Best for
direct	Direct	The official upstream link, for native behavior and the full context window
stable	Preferred	Balanced availability and speed — a good default for production traffic
economical	Economy	Cost first, well suited to batch processing and price-sensitive workloads

Available channels and their prices are listed under "Pricing" above (channels vary by model). Additional notes:

• Syntax: "provider": { "channel": "direct" }.
• If the requested channel is not enabled for that model, the request automatically falls back to the default channel and returns normally — it does not error.
• With the official SDKs: in Python pass it via extra_body; in Node.js put it directly on the request object and it passes through. In TypeScript projects, add a // @ts-expect-error line to skip the type check.

cURL

curl https://api.seawhaleai.com/v2/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <API_KEY>" \
  -d '{
    "model": "deepseek-v4-flash-vision-exp",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Hello!"}
    ],
    "provider": { "channel": "direct" },
    "stream": true
  }'
# provider is optional — remove this line to use the default channel

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.seawhaleai.com/v2",
    api_key="<API_KEY>",
)

stream = client.chat.completions.create(
    model="deepseek-v4-flash-vision-exp",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello!"},
    ],
    stream=True,
    # Optional: pick a service channel; omit to use the default
    extra_body={"provider": {"channel": "direct"}},
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

Node.js

import OpenAI from 'openai'

const client = new OpenAI({
  baseURL: 'https://api.seawhaleai.com/v2',
  apiKey: '<API_KEY>',
})

const stream = await client.chat.completions.create({
  model: 'deepseek-v4-flash-vision-exp',
  messages: [
    { role: 'system', content: 'You are a helpful assistant.' },
    { role: 'user', content: 'Hello!' },
  ],
  stream: true,
  // Optional: pick a service channel; omit to use the default
  // @ts-expect-error provider is a SeaWhale AI extension, not in the OpenAI SDK types
  provider: { channel: 'direct' },
})

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? '')
}