Xfyun
SG
Latency
-
Throughput
41tps
Uptime
Total Context
32,768
Max Output
-
Status
Available
Pricing
Input Price
Provider priced
/M tokens
Output Price
Provider priced
/M tokens
Cache Creation (5min)
-
/M tokens
Cache Read (5min)
-
/M tokens
X
Xfyun: Spark 4.0 Ultra
xfyun/spark-4.0-ultra
Current versionSpark 4.0 UltraModestable
讯飞星火旗舰非思考模型，适合结构化抽取、逻辑问答和正式办公写作
Modalities
Input
Provider priced per 1M
Output
Provider priced per 1M
context
32,768
Weekly Tokens
Live catalog
Released
Live catalog
Overview for Spark 4.0 Ultra
A compact summary of this model's best-fit workloads, routing status and production integration shape.
Architecture
Xfyun OpenAI-compatible route
Model design signal
Traffic signal
Live catalog
Recent public usage volume
Benchmark score
-
Composite coding and reasoning signal
Best for
tool usetext generation
Model tags
chatstreamingtools
Routing snapshot
Primary provider
Xfyun
Provider region
Provider policy
Current version
Spark 4.0 Ultra
Routing health
Available
Gateway reports this model ready for traffic
Xfyun
Serving traffic
Context and limits
Total Context
32,768
Max prompt
32,768
Max Output
-
Pricing tier
x1.5 gateway rate
Pricing snapshot
Input PriceProvider priced
Output PriceProvider priced
Cache creation-
Cache read-
Integration checklist
✓OpenAI-compatible chat completions
✓Call with model ID xfyun/spark-4.0-ultra
✓Use base URL https://ethergate.cn/v1
✓Create a workspace-scoped API key
Capability profile
chatStreamingTool calling
Xfyun policies apply for retention, training and regional handling.
Parameter support
max_tokenstemperaturetop_ptoolstool_choiceCheck long-context pricing once prompts exceed 32,768.
Very large prompts can add latency; test production payloads before launch.
Providers for Spark 4.0 Ultra
EtherGate routes requests to configured providers that can handle your prompt size and parameters, with fallbacks to maximize uptime. uptime.
Performance for Spark 4.0 Ultra
Compare different providers across EtherGate
Throughput
30dXfyunAverage 40 tok/s
Latency
30dXfyunAverage 1.84 s
E2E Latency
30dXfyunAverage 12.54 s
Tool Call Error Rate
30dXfyunAverage 4.83 %
Structured Output Error Rate
30dXfyunAverage 1.38 %
Effective Pricing for Spark 4.0 Ultra
Actual cost per million tokens across providers over the past hour
Weighted Average
Weighted Avg Input Price
Provider priced
Weighted Avg Output Price
Provider priced
| Provider | Input Price | Output Price | Cache Hit Rate |
|---|---|---|---|
| Xfyun | Provider priced | Provider priced | 42% |
Input Price / 1M tokens
7dMay 12May 14May 16May 18
Output Price / 1M tokens
7dMay 12May 14May 16May 18
Benchmarks for Spark 4.0 Ultra
Performance metrics and benchmarks
Ranking Distribution
1617
#1
1653
#2
1322
#3
1368
#4
Category Performance
Composite coding and reasoning signal
Score
-
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 Spark 4.0 Ultra
Top public apps this month
Available: Gateway reports this model ready for traffic
1
Hermes Agent
Open-source agent workspace for model routing
353B
Tokens
2
OpenClaw
AI coding agent connected to EtherGate
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 Spark 4.0 Ultra
Total usage per day on EtherGate
Prompt tokensReasoning tokensCompletion tokens
Uptime stats for Spark 4.0 Ultra
Uptime stats for Spark 4.0 Ultra across configured providers
Average uptime
Live
Sample code and API for Spark 4.0 Ultra
EtherGate normalizes requests and responses across providers for you.
EtherGate 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.ETHERGATE_API_KEY,
baseURL: "https://ethergate.cn/v1",
});
const completion = await client.chat.completions.create({
model: "xfyun/spark-4.0-ultra",
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://ethergate.cn/console/skills/ethergate-provider/install.sh' | bash -s -- --target qclaw --api-key '<your-ethergate-key>' --locale 'en'Required headers
Authorization
Bearer $ETHERGATE_API_KEYHeader
Content-Type: application/jsonBase URL
https://ethergate.cn/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