Latency
-
Throughput
41tps
Uptime
Total Context
1M
Max Output
384k
Status
Available
Pricing
Input Price
¥1.00 / MTok
/M tokens
Output Price
¥2.00 / MTok
/M tokens
Cache Creation (5min)
¥1.00 / MTok
/M tokens
Cache Read (5min)
¥0.02 / MTok
/M tokens
DeepSeek Official: DeepSeek V4 Flash
deepseek-official/deepseek-v4-flash
Current versionDeepSeek V4 FlashAlias targetdeepseek-v4-flashModehybrid mode
DeepSeek 官方 API 直连路由，用于与 xh.v1api.cc 对比 Token、延迟和供应商计费
Modalities
Input
¥1.00 per 1M
Output
¥2.00 per 1M
context
1M
Weekly Tokens
Comparison route
Released
Apr 24, 2026
Overview for DeepSeek V4 Flash
A compact summary of this model's best-fit workloads, routing status and production integration shape.
Architecture
284B parameter sparse MoE
Model design signal
Traffic signal
Comparison route
Recent public usage volume
Benchmark score
77.2
Composite coding and reasoning signal
Best for
cost efficiencylong contextcodingreasoning
Model tags
chatstreamingtoolsreasoninglong-contextlong context
Routing snapshot
Primary provider
DeepSeek Official
Provider region
CN
Current version
DeepSeek V4 Flash
Routing health
Missing key
Missing DEEPSEEK-OFFICIAL provider key
DeepSeek Official
Serving traffic
Context and limits
Total Context
1M
Max prompt
1M
Max Output
384k
Pricing tier
Official cache miss / cache hit
Pricing snapshot
Input Price¥1.00 / MTok
Output Price¥2.00 / MTok
Cache creation¥1.00 / MTok
Cache read¥0.02 / MTok
Integration checklist
✓OpenAI-compatible chat completions
✓Call with model ID deepseek-official/deepseek-v4-flash
✓Use base URL https://gate.topc9.com/v1
✓Create a workspace-scoped API key
Capability profile
chatStreamingTool callingReasoninglong-contextlong context
DeepSeek Official policies apply for retention, training and regional handling.
Parameter support
thinkingreasoning_effortmax_tokenstemperaturetop_pseedresponse_formattoolsCheck long-context pricing once prompts exceed 1M.
Very large prompts can add latency; test production payloads before launch.
Providers for DeepSeek V4 Flash
趣橙 routes requests to configured providers that can handle your prompt size and parameters, with fallbacks to maximize uptime. uptime.
Performance for DeepSeek V4 Flash
Compare different providers across 趣橙
Throughput
30dDeepSeek OfficialAverage 40 tok/s
Latency
30dDeepSeek OfficialAverage 1.84 s
E2E Latency
30dDeepSeek OfficialAverage 12.54 s
Tool Call Error Rate
30dDeepSeek OfficialAverage 4.83 %
Structured Output Error Rate
30dDeepSeek OfficialAverage 1.38 %
Effective Pricing for DeepSeek V4 Flash
Actual cost per million tokens across providers over the past hour
Weighted Average
Weighted Avg Input Price
¥0.8200 / MTok
Weighted Avg Output Price
¥1.72 / MTok
| Provider | Input Price | Output Price | Cache Hit Rate |
|---|---|---|---|
| DeepSeek Official | ¥1.00 / MTok | ¥2.00 / MTok | 42% |
Input Price / 1M tokens
7dMay 12May 14May 16May 18
Output Price / 1M tokens
7dMay 12May 14May 16May 18
Benchmarks for DeepSeek V4 Flash
Performance metrics and benchmarks
Ranking Distribution
1617
#1
1653
#2
1322
#3
1368
#4
Category Performance
Composite coding and reasoning signal
Score
77.2
Models Arena
3D
1297
53.0% Win · 170.0s Avg
ASCII
1196
50.0% Win · 152.0s Avg
Code Categories
1288
53.4% Win · 254.7s Avg
Data Visualization
1283
53.3% Win · 236.9s Avg
Game Development
1292
54.8% Win · 270.4s Avg
Image to HTML
1205
46.7% Win · 116.4s Avg
SVG
1239
55.2% Win · 133.4s Avg
UI Component
1303
54.5% Win · 252.7s Avg
Website
1266
51.8% Win · 306.2s Avg
Apps using DeepSeek V4 Flash
Top public apps this month
Missing key: Missing DEEPSEEK-OFFICIAL provider key
1
Hermes Agent
Open-source agent workspace for model routing
353B
Tokens
2
OpenClaw
AI coding agent connected to 趣橙
116B
Tokens
3
AA-LCR Benchmark
Applied research lab for frontier models
35.9B
Tokens
4
Claude Code Bridge
Developer tool using OpenAI-compatible APIs
26.8B
Tokens
5
Roo Code
VS Code extension for model-assisted coding
17.3B
Tokens
Monthly token usage
Apr 20May 1May 10May 18
Recent activity on DeepSeek V4 Flash
Total usage per day on 趣橙
Prompt tokensReasoning tokensCompletion tokens
Uptime stats for DeepSeek V4 Flash
Uptime stats for DeepSeek V4 Flash across configured providers
Average uptime
Official status
Sample code and API for DeepSeek V4 Flash
趣橙 normalizes requests and responses across providers for you.
趣橙 supports reasoning-enabled models that can return step-by-step reasoning metadata. Use the reasoning parameter when a provider supports it, and preserve reasoning_details when continuing a conversation. Learn more about reasoning tokens.
Request
Use the OpenAI-compatible Chat Completions endpoint. Keep your SDK, change the base URL, API key and model ID.
OpenAI SDK
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.QICHENG_API_KEY,
baseURL: "https://gate.topc9.com/v1",
});
const completion = await client.chat.completions.create({
model: "deepseek-official/deepseek-v4-flash",
messages: [{ role: "user", content: "Explain your best use case." }],
});Tool configuration examples
Use this model ID directly in agent tools that accept OpenAI-compatible endpoints.
QClaw macOS/Linux
curl -fsSL 'https://gate.topc9.com/console/skills/qicheng-provider/install.sh' | bash -s -- --target qclaw --api-key '<your-qicheng-key>' --locale 'en'Required headers
Authorization
Bearer $QICHENG_API_KEYHeader
Content-Type: application/jsonBase URL
https://gate.topc9.com/v1Optional headers
Optional attribution headers help you identify traffic by app or website when debugging usage.
HTTP-Refererhttps://your-app.example
X-TitleYour app name
Using third-party SDKs
openai-js
openai-python
Vercel AI SDK
LangChain