DeepSeek

deepseek-v4-flash

Model IDdeepseek-v4-flash

Integration docsMarkdown

Model introduction

DeepSeek V4-Flash is the fast, economical open-weight V4 model. It supports a 1M context window, thinking and non-thinking modes, and OpenAI- and Anthropic-compatible APIs.

Model capabilities

Specifications come from the cited model material. Catalog tags help identify access features; use the source documentation for exact limits.

OpenAI compatibleResponses APIAnthropic compatibleGemini compatiblePrompt cacheReasoning

Verified specifications

Official positioning
Fast, efficient, economical V4 model
Context window
1,000,000 tokens
Maximum output
384,000 tokens
Model size
284B total / 13B active parameters

Sources: DeepSeek models and pricing documentation

Independent capability evaluation

Only public measurements matched to the exact model ID and reasoning profile are shown; nearby variants are not substituted.

Intelligence index

40

Cohort rank #11 / 97

Output speed

120.4 tok/s

Cohort rank #12 / 97

First-token latency

1.26s

Reasoning, max effort

Trades a modest composite-score gap for much higher speed and lower price, making it a throughput-first open-weight option.

Evaluation profile: Reasoning, max effort. Artificial Analysis · Methodology

Model pricing and access

Prices are read directly from the MoleAPI console catalog and shown by current billing group and context tier.

Your final charge follows the account group shown in the console.
Live pricing source

Pricing table

USD / 1M tokens
Standardx1defaultInput $0.22 · Output $0.67

Input

$0.22 / 1M tokens

Output

$0.67 / 1M tokens

Cache read

$0.0075 / 1M tokens

Cache write

$0.275 / 1M tokens

Discountx0.8discountInput $0.176 · Output $0.536

Input

$0.176 / 1M tokens

Output

$0.536 / 1M tokens

Cache read

$0.006 / 1M tokens

Cache write

$0.22 / 1M tokens

Relayx0.3relayInput $0.066 · Output $0.201

Input

$0.066 / 1M tokens

Output

$0.201 / 1M tokens

Cache read

$0.0023 / 1M tokens

Cache write

$0.0825 / 1M tokens

Temporary (availability may vary)x0.1tempInput $0.022 · Output $0.067

Input

$0.022 / 1M tokens

Output

$0.067 / 1M tokens

Cache read

$0.0008 / 1M tokens

Cache write

$0.0275 / 1M tokens

Access protocols

Available billing groups: Standard, Discount, Relay, Temporary (availability may vary)

openai
POST
/v1/chat/completions
openai-response
POST
/v1/responses
anthropic
POST
/v1/messages
gemini
POST
/v1beta/models/{model}:generateContent
openai-response-compact
POST
/v1/responses/compact
openai-alpha-search
POST
/v1/alpha/search

Complete model guide

Prepared from vendor sources and independently reviewed. · 2026-07-28

DeepSeek-V4-Flash is an API model in the DeepSeek V4 series. It supports thinking and non-thinking modes and provides capabilities including long context, JSON Output, tool calling, chat prefix continuation, and FIM Completion.[4]

Model Positioning and Core Capabilities

DeepSeek-V4-Flash is designed for text generation workflows that can switch between thinking and non-thinking modes. Thinking mode is enabled by default.[4] The model supports JSON Output, Tool Calls, chat prefix continuation (Beta), and FIM Completion (Beta).[4]

Thinking mode can be used for generation tasks involving reasoning processes, while non-thinking mode provides an alternative generation mode.[4] DeepSeek describes V4 Preview as having stronger Agent capabilities and reasoning capabilities, and it is available on the web, in the App, and through the API.[6]

Capabilities and Specifications

DeepSeek-V4-Flash provides a 1M context window and a maximum output length of 384K, making it suitable for tasks that process long inputs and generate longer results.[4] The officially listed concurrency limit is 2500.[4]

SpecificationParameter
Context length1M tokens [4]
Maximum output length384K tokens [4]
Thinking modeSupported, enabled by default [4]
Non-thinking modeSupported [4]
JSON OutputSupported [4]
Tool CallsSupported [4]
Chat prefix continuationSupported, Beta [4]
FIM CompletionSupported, Beta; non-thinking mode only [4]
Concurrency limit2500 [4]

Use Cases

The 1M context window of DeepSeek-V4-Flash supports long-document processing, multi-turn conversations, and text generation workflows that need to retain substantial context.[4] JSON Output and Tool Calls can be used for structured result generation and tool-calling workflows.[4]

Chat prefix continuation is suitable for workflows that continue generating from existing text.[4] FIM Completion can be used for scenarios that require completing content within existing text, but this capability is available only in non-thinking mode.[4]

Important Limitations

FIM Completion supports only non-thinking mode and cannot be used together with thinking mode.[4] Both chat prefix continuation and FIM Completion are labeled as Beta features.[4]

The model concurrency limit is 2500. Rate limiting and isolation rules when this limit is exceeded are defined by DeepSeek's applicable service rules.[4]

Sources

[4] DeepSeek 模型 & 价格文档 [6] DeepSeek 官网

Code examples

These examples use MoleAPI's Responses API endpoint and run after you replace the API key.

curl https://api.moleapi.com/v1/responses \
  -H "Authorization: Bearer $MOLEAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "input": "Explain this problem step by step."
  }'

Frequently asked questions

How is deepseek-v4-flash priced?

This page reads prices from the MoleAPI console API and updates with model prices, context tiers, and account groups.

How can I access deepseek-v4-flash?

Protocols and endpoints come from the supported_endpoint_types field in the MoleAPI model catalog.

How do I switch an existing project to deepseek-v4-flash?

Keep the MoleAPI API address and key, replace the model parameter with the model ID on this page, then check protocol-specific parameter differences.

Where do the deepseek-v4-flash model details come from?

Capabilities and limitations are checked against DeepSeek and the other cited pages. Pricing and available protocols come only from the MoleAPI console.