deepseek-flash

deepseek-flash

DeepSeek’s newly released language model, designed for high-performance production scenarios, is specifically optimized for ultimate inference efficiency and response speed.
2026-09-08
LLM
Model capability: thinkingModel capability: function_call
Cache Read:
$0.003/1M tokensstarting from
Input:
$0.15/1M tokensstarting from
Output:
$0.6/1M tokensstarting from
Bulk order? Contact your manager for exclusive deals

API Overview

Please translate the following content into English, returning only the translation without any additional explanations:

DeepSeek Flash is DeepSeek’s ultra-lightweight flagship large model, designed for massive high-concurrency, low-latency, and cost-effective code and data processing. While maintaining exceptional reasoning capabilities and rigorous logic, DeepSeek Flash compresses generation latency to the millisecond level and reduces overall API costs to an extremely low level. The model natively supports ultra-long context and a deep prompt caching mechanism, making it highly suitable as a cost-effective industrial-grade foundation for high-frequency customer service, complex workflow routing, large-scale log/data extraction and cleansing, and code assistant sub-agents.

───────────────────────────────────────────────────────────────────

Core Capabilities

Ultra-High Throughput with Millisecond Latency: Extremely fast first-token response time (TTFT) and high tokens-per-second (TPS), perfectly suited for high-concurrency and real-time interaction scenarios.

Deep Prompt Caching Compute Optimization: Natively supports a very high context cache hit discount, significantly reducing the cost of repetitive input in long conversations, frequent system prompt attachments, or multi-turn agent loops.

Strong Logic and Instruction Following: Inherits the DeepSeek family’s strong genes in mathematics, logical reasoning, and code generation, with excellent adherence to structured JSON output and complex tool calling.

Time-Based Tiered Pricing: Natively supports differentiated pricing during idle and peak hours, enabling enterprises to schedule massive asynchronous batch tasks to off-peak windows to lower costs.

Playground

Log in to explore more features! Click to Log In

API Analytics

API Reference (3)

API DescriptionAPI EndpointRequest MethodStabilityParameter Description
Chat (DeepSeek)
POST
Stable
View Details
Chat (DeepSeek)
POST
Stable
View Details
Messages (for Claude Code)
POST
Stable
View Details

API Pricing

$¥ 円 ₽
ModelDescriptionContextOfficial Price302.AI PriceOfficial Price Gap

deepseek-chat

Off-peak Pricing
1000000

Cache Read$0.003 / 1M tokens
Input$0.15 / 1M tokens
Output$0.6 / 1M tokens

Cache Read$0.003 / 1M tokens
Input$0.15 / 1M tokens
Output$0.6 / 1M tokens

Original Price

deepseek-chat

Peak Pricing (UTC: 01:00‑04:00,06:00‑10:00)
1000000

Cache Read$0.006 / 1M tokens
Input$0.3 / 1M tokens
Output$1.2 / 1M tokens

Cache Read$0.006 / 1M tokens
Input$0.3 / 1M tokens
Output$1.2 / 1M tokens

Original Price