DeepSeek V4 Flash

DeepSeek V4 Flash for low-latency agent, coding and high-volume production workloads

Published

scheduleReleased：July 31, 2026

DeepSeek V4 Flash is a DeepSeek V4 model currently backed by DeepSeek-V4-Flash-0731, with a 1M context window, up to 384K output, and thinking and non-thinking modes.

starsCapabilities

codeFunction callingdata_objectStructured output

paymentsContext and pricing

Context limit1,000,000

Max output384,000

Input price¥3/ 1M tokens

Output price¥9/ 1M tokens

Cached input price¥0.1/ 1M tokens

descriptionOverview

Overview

DeepSeek V4 Flash is a DeepSeek V4 model with encyclopedia identifier deepseek-v4-flash, currently backed by DeepSeek-V4-Flash-0731. It supports a 1M context window, up to 384K output, thinking and non-thinking modes, Tool Calls, JSON Output, the Responses API and the Anthropic API.

Versioning

deepseek-v4-flash and deepseek-v4-pro are the official stable API names and track the current versions. Dated entries are encyclopedia snapshots only; DeepSeek does not document them as callable API model parameters. The legacy deepseek-chat and deepseek-reasoner names were retired on July 24, 2026.

lightbulbUse cases

Agents and tool use
Coding and repository tasks
Long-context analysis
Chinese Q&A and content generation

thumb_upStrengths

1M context window
Up to 384K output
Thinking and non-thinking modes
OpenAI and Anthropic API compatibility

infoLimitations

Stable API names track newer official versions
Pinned evaluations should record dated versions
Latency and cost require workload testing
Official API pricing may increase soon