DeepSeek V4 Flash

Lightweight MoE model with 284B total parameters and 1M context window, delivering near-flagship performance at a fraction of the cost.

Overview

DeepSeek-V4-Flash is the lightweight variant of the DeepSeek-V4 series, a Mixture-of-Experts (MoE) model with 284B total parameters and 13B activated parameters. It delivers reasoning capabilities that closely approach the larger V4-Pro model, making it an excellent choice for cost-sensitive applications that still require strong performance.

Key Specifications

Parameter	Value
Context Window	1,000,000 tokens (1M)
Max Output	384,000 tokens (384K)
Total Parameters	284B
Activated Parameters	13B
Architecture	MoE (Mixture-of-Experts)
License	MIT

Capabilities

Text Generation — High-quality natural language for conversation, analysis, and creative writing
Code Generation — Strong coding performance across multiple languages
Function Calling — Structured tool use for API integrations
JSON Mode — Guaranteed valid JSON output
Thinking Mode — Supports non-thinking, think-high, and think-max reasoning levels
Streaming — Server-sent events for real-time token delivery
Chat Prefix Completion — Beta support for agent-style workflows
FIM (Fill-In-the-Middle) — Beta support for code completion

Pricing

Type	Price
Input	$0.14 / 1M tokens
Output	$0.28 / 1M tokens
Cache Read	$0.0028 / 1M tokens

Usage Example

from openai import OpenAI

client = OpenAI(
    api_key="sk-...",
    base_url="https://api.yolorouter.com/v1",
)

response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Hello!"}],
)

print(response.choices[0].message.content)

Best Practices

Use non-thinking mode for routine tasks (simple chat, low-risk decisions) to minimize latency and cost
Use think-high mode for complex problem-solving, coding, and planning
Leverage the 1M context window with cache-friendly prompt patterns to maximize cache hit rate