XIANSHE AI · UNIFIED MODEL WORKSPACE

One API.
Leading AI models.

Access text, image, video and speech models with one Xianshe account. Membership, points, community compute and cloud providers work together.

Create API Key

Membership recognized Unified points billing Automatic failure refunds Provider load balancing

API service operational V2.4.6

闲

Current accountGuest mode

Guest

0Available points

0Available tokens

18Text models

4Multimodal models

Task routing

Community computePreferred

Cloud providersAutomatic failover

AI CREATION STUDIO

Developer Access

Operational · Keys encrypted server-side

闲

Xianshe AgentTask-focused AI assistant

Select a text model

Quick start

闲

闲社智能体

告诉我你想完成什么。我会根据任务选择聚合模型，并统一使用你的会员权限与积分。

0 / 8000 · Enter to send

TEXT TO IMAGE

Turn an idea into an image

For posters, product visuals, covers, social graphics and concept art.

Image prompt

ModelSize

ImgYour image will appear hereFailed generations do not deduct points

TEXT TO VIDEO

Create short videos and motion assets

Generation runs in the background. Task status updates automatically.

Video prompt

Aspect ratioResolutionDuration

VidNo video task yetGeneration typically takes tens of seconds

TEXT TO SPEECH

Text to Speech

Generate speech for narration, presentations and short-video voiceovers.

Speech text Voice

SPEECH TO TEXT

Speech to Text

Upload a short audio file or enter a public audio URL for transcription.

Upload audio Or enter an audio URL

识别结果将在这里显示。

OPENAI COMPATIBLE

One key for Xianshe AI services

For agents, workflows, desktop clients and your own apps. Upstream provider keys stay on the server.

Base URLhttps://www.xianshe.com/source/plugin/xianshe_ai/openai.php/v1

chat-completions.json

API capabilities

POST/chat/completionsText chat

POST/images/generationsImage Generation

POST/videos/generationsVideo Generation

GET/videos/tasks/task_idVideo status

POST/audio/speechSpeech synthesis

POST/audio/transcriptionsSpeech recognition

MEMBERSHIP & BILLING

One account, unified management

Points are the billing currency. Tokens measure model usage.

MODEL ROUTER

Available Models and Services

Models from disabled providers are hidden and excluded from routing.

AIMiniMax-M2.5MiniMax-M2.5Available

AIdeepseek-v3.2deepseek-v3.2Available

AIdeepseek-v4-flashdeepseek-v4-flashAvailable

AIdeepseek-v4-flash-0731deepseek-v4-flash-0731Available

AIdeepseek-v4-prodeepseek-v4-proAvailable

AIdeepseek-v4-pro-0813deepseek-v4-pro-0813Available

AIglm-5glm-5Available

AIglm-5.1glm-5.1Available

AIglm-5.2glm-5.2Available

AIkimi-k2.5kimi-k2.5Available

AIkimi-k2.6kimi-k2.6Available

AIkimi-k2.7-codekimi-k2.7-codeAvailable

AIqwen3.6-flashqwen3.6-flashAvailable

AIqwen3.6-plusqwen3.6-plusAvailable

AIqwen3.7-maxqwen3.7-maxAvailable

AIqwen3.7-plusqwen3.7-plusAvailable

AIqwen3.8-flashqwen3.8-flashAvailable

AIqwen3.8-maxqwen3.8-maxAvailable

ImgWan Image 3.0 Proqwen-image-3.0-proImage generation · points/request

VidBailian HappyHorse 1.1 Videohappyhorse-1.1-t2vVideo generation · points/request

AudQwen TTS Plusqwen-audio-3.0-tts-plusSpeech synthesis · points/request

ASRQwen ASR Flashqwen-audio-3.0-asr-flashSpeech recognition · points/request

PPrompt LibraryUse expert prompts with AI models GPUCompute Center0 nodes · Earnings ¥0 WWorkflow MarketInstall complete, executable workflows

Your First API Request: Preparation and Verification

Sign in to Xianshe, create your own API key on this page, and check permissions, balance and current fees. Do not use upstream provider keys.

Copy the Base URL and an available text model ID. Display names may differ; use the model ID.

Set XIANSHE_AI_API_KEY and XIANSHE_AI_MODEL in your environment. Keep keys out of webpages, screenshots and public repositories.

After checking fees, send a short request without sensitive data. Verify the response, usage and points ledger.

Python 3 Text Request Example

This example uses the Python standard library and a 32-token output limit. Configure your key and an available model ID, then check trial eligibility and fees. The actual request and billing evidence below was recorded with a separate PHP cURL client.

import json import os import urllib.request base_url = "https://www.xianshe.com/source/plugin/xianshe_ai/openai.php/v1" payload = { "model": os.environ["XIANSHE_AI_MODEL"], "messages": [{"role": "user", "content": "请只回复：连接成功"}], "max_tokens": 32, "stream": False } request = urllib.request.Request( base_url + "/chat/completions", data=json.dumps(payload, ensure_ascii=False).encode("utf-8"), headers={ "Authorization": "Bearer " + os.environ["XIANSHE_AI_API_KEY"], "Content-Type": "application/json" }, method="POST" ) with urllib.request.urlopen(request, timeout=90) as response: result = json.load(response) print(result["choices"][0]["message"]["content"]) print(result["usage"])

Public API: A Recorded Connectivity Check

On October 20, 2026, an independent client completed a request using an existing Xianshe key and the public HTTPS endpoint. The internal acceptance record below is provided for comparison.

Request method: POST /chat/completions, Bearer key authentication, streaming disabled
Recorded model ID: qwen3.8-flash; use a model ID currently available to your account
Actual input (Chinese): 请只回复：连接成功
Output limit: 32 Token
Actual response (Chinese): 连接成功; HTTP 200, finish_reason: stop
Recorded usage and billing: 40 input tokens and 3 output tokens, 43 in total. This request used 43 trial tokens, deducted 0 points and released the reservation.

The zero-point charge used the test account's trial allowance at that time. Your current account and model rules determine eligibility and fees. This record verifies authentication, request, response and billing, not performance or customer delivery.

The Python example uses the same request fields and 32-token output limit for reproduction in your environment. The recorded client used PHP cURL. Check your available models and fees before running it.

View Redacted Response and Billing Record (JSON)

Frequently Asked Questions

How do I verify a successful integration?

Confirm that the selected model returns content and the corresponding usage record is present. Opening a page, copying a key or receiving an HTTP response alone does not confirm a completed model task.

What should I check after an authentication or model error?

Check key validity, completeness and account permissions for authentication errors, and the available model ID for model errors. For network failures, inspect task and usage records before retrying a paid request.

Does this example also cover images, video and speech?

This example covers text chat only. Other capabilities use different endpoints and parameters. Follow their API and model requirements; video also requires task-status queries.