FAQ

AIPower is an OpenAI-compatible API gateway providing access to 16 AI models from 10 providers through one endpoint. Includes GPT-5.4, Claude, Gemini, and 10 Chinese AI models (DeepSeek, Qwen, GLM, Kimi, Doubao, MiniMax) — the only gateway with this breadth of Chinese AI.

Sign up free at aipower.me (no credit card). You get 50 free API calls immediately. Use your API key with any OpenAI SDK by changing base_url to https://api.aipower.me/v1.

No — just use the standard openai Python/Node/Go SDK. Change two lines (base_url and api_key). That's it.

Yes. Visit aipower.me/try — 5 free messages per day per IP, no registration. For more, sign up free.

Pay-per-token, no subscriptions. Models range from $0.01/M tokens (GLM-4 Flash) to $19.50/M tokens (GPT-5.4 / Claude Sonnet output). See full pricing at aipower.me/pricing.

Credit card (USD via Stripe), WeChat Pay (CNY or HKD), and Alipay. Minimum top-up is $5.

No. Your pre-paid credits never expire.

No subscriptions. Pay only for what you use. Add credits when needed.

Stack: Sign up (50) → First API call (+10) → Add payment method (+50) → First top-up $5+ (+100) → Refer a friend (+200/+500). Max 310+ free calls before paying.

16 models: GPT-5.4, GPT-4o Mini, Claude Opus 4.6, Claude Sonnet 4, Gemini 2.5 Pro/Flash, DeepSeek V3/R1, Qwen Plus/Turbo, GLM-5.1, GLM-4 Flash, Kimi K2.5, Moonshot v1 8K, MiniMax Text 01, Doubao Pro 256K.

Use model="auto" and AIPower picks the best model for your task. Options: auto (balanced), auto-cheap (cheapest), auto-best (most powerful), auto-code (best coder), auto-fast, auto-free.

Yes. AIPower handles Chinese provider access for you — no +86 phone, no Chinese bank card needed. Just sign up with email.

Yes. AIPower is reachable from within China (via Cloudflare). Our servers handle the upstream connection to Anthropic/OpenAI/Google — your IP never connects to them directly.

Claude Sonnet 4 (HumanEval 93.7%) and GLM-5.1 (92.1%) are tied top. DeepSeek V3 (92.7%) is 13x cheaper than Claude and excellent for most tasks. Try them at /compare.

GLM-4 Flash at $0.01/M tokens (both input and output). Perfect for classification, extraction, or high-volume tasks.

Yes. Drop-in replacement. Python, Node.js, Go, curl — any OpenAI client works. Change base_url to https://api.aipower.me/v1 and use your AIPower key.

Yes. Server-Sent Events (SSE) streaming supported for all chat completion endpoints.

Yes — supported on models that support it (GPT-5.4, Claude Sonnet/Opus). Works with LangChain, Vercel AI SDK, and raw OpenAI SDK.

Free tier: 5 requests/minute. Paid tier: 60 requests/minute default. Higher limits for enterprise — email enterprise@aipower.me.

Under 50ms routing overhead (Cloudflare edge). Total latency = our routing + upstream model latency. Typically 200-500ms from Asia, 100-300ms from US.

Yes. /v1/responses endpoint is supported alongside /v1/chat/completions.

Embeddings support is coming soon. For now, use chat completions for text understanding tasks, or email us for early access.

No. We don't store prompts or responses. We only log token counts for billing. All traffic is TLS-encrypted.

Keys are SHA-256 hashed in our database. Original key is shown only once at creation. Revoke/rotate at /dashboard/keys.

SOC 2 is in progress. Current controls: Stripe-level payment security, Cloudflare DDoS protection, no prompt logging, TLS everywhere. DPA available for enterprise.

Cloudflare Workers run at the edge closest to you. Upstream calls go to provider regions (OpenAI: US, Anthropic: US, DeepSeek/Qwen: China). We don't route through intermediaries.

可以。api.aipower.me 在 Cloudflare 上，国内直连，不需要翻墙。

支持。可以用微信支付或支付宝以 CNY 或 HKD 充值，最低 $5（约 ¥36）。

不会。您的 IP 只连接到我们的 Cloudflare 节点，不会直接接触 OpenAI/Anthropic。我们的上游账号是企业级账号，稳定性有保障。

企业客户可开 AI POWER LIMITED (香港公司) 的商业发票。联系 frank@aipower.me。

We focus on Chinese AI model access (10 models vs OpenRouter's few), smart routing (6 modes), WeChat Pay, and a free tier with no card. OpenRouter has broader model variety (300+ niche models). We're complementary — many devs use both.

Together AI focuses on open-source model hosting. AIPower is a unified gateway to proprietary and Chinese models. Use Together for custom Llama/Mistral deployments, AIPower for GPT+Claude+DeepSeek via one API.

OpenAI is one provider. AIPower gives 16 models — use GPT for complex reasoning, DeepSeek V3 for general tasks at 91% lower cost, GLM-4 Flash for nearly-free classification. Smart routing saves 80-90% on total bills.

Yes. Volume discounts (up to 30%), dedicated support, SSO, custom contracts, 99.9% SLA. See aipower.me/enterprise or email enterprise@aipower.me.

Yes for enterprise plans. Send a PO or signed agreement to enterprise@aipower.me.

Not currently. Our gateway runs on Cloudflare Workers. For air-gapped deployments, contact enterprise@aipower.me to discuss.

Yes — standard GDPR-compliant Data Processing Agreement available on request.