Unified API access to DeepSeek and other leading AI models. OpenAI-compatible interface, pay-as-you-go pricing, international bank transfer payment. No credit card required.
Yingsuan AI (盈算智服) provides unified API access to AI models and GPU computing resources for developers and teams. Our platform offers OpenAI-compatible API endpoints for DeepSeek models, along with GPU compute rental services (H100, A100, V100, and more). Pay only for what you use — no subscriptions, no minimum commitments.
Released July 31, 2026. Agent & coding ability jumped 7× on SWE tasks. Cache-hit input costs just 1/50 of regular input — multi-turn agents and long conversations save 90%+. Native Responses API and Anthropic-format endpoints mean your existing OpenAI or Claude workflow migrates with a one-line base-URL change. Already enabled on every Yingsuan key — including the 50-request Promo free trial.
Open weights released by Moonshot AI on July 27, 2026. Use K3 instantly through the Yingsuan gateway — no GPU setup required.
No credit card required. 100 free requests on 3 free models. Test DeepSeek, GLM, Qwen instantly.
Built for developers, optimized for real-world workloads
YingSuan is an independent platform providing API access to AI models. We are not affiliated with, endorsed by, or connected to DeepSeek, OpenAI, or any other AI model provider. Our service provides a unified API gateway for accessing third-party AI models with consistent formatting and enhanced availability for developers worldwide.
Per 1 million tokens. Real-time rates from /v1/pricing (no key required).
| Model | Input (¥ / 1M tokens) | Output (¥ / 1M tokens) | Description |
|---|---|---|---|
| deepseek-v4-flash NEW | $0.20 off-peak | $0.39 off-peak | DeepSeek-V4-Flash-0731. 1M context, 384K max output. Top-tier agent & coding (DeepSWE 54.4, Terminal-Bench 82.7). Cache-hit input just $0.004/1M (regular input 1/50 — multi-turn agents save 90%+). Free 50-request Promo trial available. Legacy names deepseek-chat and deepseek-reasoner still route here — no code change needed. |
| deepseek-v4-pro | $0.59 off-peak | $1.17 off-peak | DeepSeek-V4-Pro. 1M context. For heavy reasoning and long-horizon agent runs. |
| qwen3.6-flash ✅ 已接入 | $0.30 | $0.90 | Alibaba Qwen3.6-Flash. 1M context. Fast & versatile. Lightweight, cost-efficient. |
| qwen3.7-plus ✅ 已接入 | $0.60 | $1.80 | Alibaba Qwen3.7-Plus. 1M context. Balanced performance for most tasks. |
| qwen3.7-max ✅ 已接入 | $3.00 | $9.00 | Alibaba Qwen3.7-Max. 1M context. Qwen flagship. Strongest capability. |
| qwen3.8-flash 🆕 新开源 | $0.21 | $0.63 | Alibaba Qwen3.8-Flash (open-sourced 2026-08-26). 1M context. New QSA/GDN arch — outperforms Claude Opus 4.6, close to 4.8. Native multimodal understanding (image/doc input). Top-tier agent & coding. Inference cost ≈ 1/5 of V4-Flash. |
| glm-4-flash FREE | $0.00 | $0.00 | Zhipu GLM-4-Flash. 128K context. Free (Zhipu official) — recommended stable starting model. |
| glm-4.7-flash FREE | $0.00 | $0.00 | GLM-4.7-Flash. 200K context. Coding-focused (Zhipu rate-limits more often — for a stable start use glm-4-flash). Free (Zhipu official). |
| glm-4-air 1M FREE | $0.12 | $0.30 | Zhipu GLM-4-Air. 128K context. 1M free tokens/month. High cost-efficiency. |
| glm-4-plus | $0.75 | $1.50 | Zhipu GLM-4-Plus flagship. 128K context. Complex reasoning. |
| glm-5.2 NEW | $1.20 | $3.90 | Zhipu GLM-5.2 flagship (NEW). 1M context. Coding SOTA (Code Arena #1). Default deep-thinking on — raise max_tokens ≥ 1024 per our Thinking-Models guide. |
| glm-5.3-flash 🆕 新开源 | $0.21 | $0.63 | Zhipu GLM-5.3-Flash (open-sourced 2026-08-26). 1M context. GLM-5 series' first native multimodal. Artificial Analysis index 57 = Claude Opus 4.8. China-hosted compute. Coding/Agent strong. Price ≈ 1/10 of GLM-5.2. |
| kimi-k3 🔥 开源权重已开放 | $3.33 | $16.67 | Moonshot Kimi K3 — 全球最大开源 2.8T 模型，100万 token 上下文，前端代码竞技场 #1（超 Claude Fable 5），AA 智能指数第 4。缓存命中输入仅 $0.33/M。⚠️ 默认全力思考，输出较长、幻觉率偏高，事实/研究类任务建议加验证。 |
| MiniMax-M3 NEW | $0.60 | $2.40 | MiniMax M3. 1M context. Native multimodal (image/video input). Built for coding, tool-use agents & long docs. OpenAI & Anthropic compatible. Adaptive thinking on — raise max_tokens ≥ 1024. |
| Qwen/Qwen2.5-7B-Instruct FREE | $0.00 | $0.00 | SiliconFlow. Qwen2.5 7B open-source. Free (SiliconFlow official). 32K context. |
| Qwen/Qwen2.5-72B-Instruct | $0.84 | $1.69 | SiliconFlow. Qwen2.5 72B open-source. 32K context. Strong reasoning. |
| deepseek-ai/DeepSeek-V3 | $0.56 | $1.13 | SiliconFlow. DeepSeek V3 via SF. 64K context. Alternative endpoint. |
| deepseek-v4-flash-vision-exp 👁 视觉理解 | $0.20 off-peak | $0.39 off-peak | DeepSeek V4-Flash 视觉理解实验版 · 图文输入→文本输出 · 1M 上下文 · 看图识别/读图/单据结构化（汽修：拍维修单/故障件/车牌→结构化）· 价格与 V4-Flash 一致 · 实验性模型，生产路径建议配置 fallback。 |
| qwen3-vl-flash 👁 视觉理解 | $0.80 | $2.40 | 通义千问 Qwen3-VL-Flash 视觉理解 · 图文输入→文本输出 · 128K 上下文 · 中文 OCR/车牌/VIN/单据识别强 · 汽修拍照识别增强。非图像/视频生成。 |
| qwen-vl-ocr 👁 高精中文OCR | $0.80 | $2.40 | 通义千问专用中文 OCR · 高精度识别车牌/VIN/发票/工单/铭牌文字 · 汽修「高精中文识别」场景引擎。图文输入→文本输出，非图像/视频生成。 |
| MiniMax-M2.7 | $0.30 | $1.20 | MiniMax M2.7. 200K context. 递归自我改进，高性价比。T1 标准档轻量首选。 |
Select your task type and budget. The guide reads live gateway pricing and peak/off-peak windows to recommend the most cost-effective model for your workload — including DeepSeek, Kimi K3, GLM, Qwen, and MiniMax.
💡 1M tokens ≈ 750,000 English words ≈ 500,000 Chinese characters
💡 Example: A typical chat conversation of 20 messages uses ~5,000 tokens, costing less than $0.01
💡 GPU compute pricing: See GPU pricing comparison
Get running in 3 simple steps
# Chat completion request curl https://yingsuan.top/v1/chat/completions \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "deepseek-v4-flash", "messages": [ {"role": "user", "content": "Hello! What can you do?"} ], "max_tokens": 100 }' # List available models curl https://yingsuan.top/v1/models \ -H "Authorization: Bearer YOUR_API_KEY"
from openai import OpenAI client = OpenAI( api_key="YOUR_API_KEY", base_url="https://yingsuan.top/v1" ) response = client.chat.completions.create( model="deepseek-v4-flash", messages=[ {"role": "user", "content": "Hello! What can you do?"} ], max_tokens=100 ) print(response.choices[0].message.content)
import OpenAI from 'openai'; const client = new OpenAI({ apiKey: 'YOUR_API_KEY', baseURL: 'https://yingsuan.top/v1', }); const response = await client.chat.completions.create({ model: 'deepseek-v4-flash', messages: [ { role: 'user', content: 'Hello! What can you do?' } ], max_tokens: 100, }); console.log(response.choices[0].message.content);
# For ChatBox, OpenWebUI, LibreChat, and other OpenAI-compatible apps: API Base URL: https://yingsuan.top/v1 API Key: YOUR_API_KEY Model Name: deepseek-v4-flash (or deepseek-chat) # Just paste these 3 values into your app's settings, # and you're ready to go!
OpenAI-compatible REST API
| model | string, required | "deepseek-v4-flash" or "deepseek-chat" |
| messages | array, required | Array of {role, content} objects |
| max_tokens | integer, optional | Maximum tokens to generate (default: 4096) |
| temperature | float, optional | 0.0 - 2.0 (default: 1.0) |
| stream | boolean, optional | Enable streaming responses (default: false) |
| Authorization | header, required | "Bearer YOUR_API_KEY" |
Choose a plan below and email us to order. Pay via international bank transfer (Wise USD). No credit card required.