One API key for
every frontier model
Natively compatible with the OpenAI and Anthropic SDKs. Point your base URL at us and keep the rest of your code exactly as it is — calling Claude, GPT and Gemini for as little as 7% of list price.
Free trial credit on signup · No credit card required
curl https://apihub.ltd/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-5",
"messages": [{"role": "user", "content": "Hello!"}],
"stream": true
}'- 99.9%
- Uptime
- 0.4s
- Avg. time to first token
- 120+
- Models available
- 32M
- Requests per day
Every major provider, kept in sync as new models ship.
Why teams choose us
From unit economics to uptime, every part of a gateway done properly.
Dramatically lower cost
Consistently as low as 7% of official pricing. Billed per token, with no minimum spend and no hidden fees — you pay for exactly what you use.
Drop-in compatible
Full support for the OpenAI and Anthropic wire protocols. Swap the base URL and the key, and your existing code just works.
Multi-route high availability
Traffic is balanced across multiple upstream channels with automatic failover, holding uptime above 99.9% even when a provider degrades.
Every major model
120+ models from Anthropic, OpenAI, Google, xAI and DeepSeek behind one endpoint, with new releases enabled as they ship.
Usage you can actually see
Every call logged with model, token counts, latency and cost. Break usage down per key and export the raw records whenever you need them.
Enterprise-grade security
TLS end to end and zero retention of prompt or completion content. Set per-key spend limits, model allowlists and expiry dates.
Live in three steps
No business logic to rewrite. Migrate in about a minute.
- 01
Create a key
Sign up, open the console and create an API key. Give it a name, a spend limit and an allowlist of models if you want one.
- 02
Swap the base URL
Replace the official endpoint with ours and the official key with the one you just created. Everything else stays untouched.
- 03
Start calling
Send requests as usual. Usage, spend and per-call details stream into your console in real time.
Supported models
Every major provider, kept in sync as new models ship.
| Model | Context | Input / 1M | Output / 1M |
|---|---|---|---|
Claude Opus 5 claude-opus-5 | 200K | $1.05$15.00 | $5.25$75.00 |
Claude Sonnet 5 claude-sonnet-5 | 200K | $0.300$3.00 | $1.50$15.00 |
Claude Opus 4.8 claude-opus-4-8 | 200K | $1.05$15.00 | $5.25$75.00 |
GPT-5.2 gpt-5.2 | 400K | $0.150$1.25 | $1.20$10.00 |
Gemini 3 Pro gemini-3-pro | 1M | $0.150$1.25 | $1.20$10.00 |
Billing that fits your usage
Pay as you go or subscribe — pick whichever matches how you actually call.
Pay as you go
Most flexibleTop up once, then pay per token consumed. Credit never expires and you only spend what you use.
- Access to every model
- Credit never expires
- Precise per-token billing
- No minimum spend
- Independent limits per key
Subscription
Best for scaleA fixed monthly allowance at a lower unit price — built for steady production traffic.
- Fixed monthly allowance
- Lower effective unit price
- Higher rate and concurrency limits
- Priority support channel
- Falls back to pay-as-you-go
Frequently asked questions
Still unsure about something? Get in touch any time.
Do I have to change my existing code?
Barely. We implement the OpenAI and Anthropic wire protocols in full, so you only point the SDK's base URL at our gateway and swap in a key from your console. Request parameters, response shapes and streaming behaviour are all unchanged.
How can the pricing be so much lower than official?
We buy capacity at scale across multiple channels and pass most of that saving on. Billing works exactly like the official one — input and output tokens are priced separately — and you can check the current unit price for every model on the pricing page.
Does credit expire? Is there a minimum spend?
Pay-as-you-go credit never expires, and there is no minimum spend or monthly fee of any kind. Subscription allowances reset on a calendar-month cycle, and anything beyond the allowance can fall back to pay-as-you-go automatically.
Do you store my prompts or completions?
No. We keep only the metadata required for billing and debugging — model name, token counts, latency and status code. Request bodies and model output are never written to disk, and everything is encrypted in transit.
Are streaming and tool use supported?
Yes. SSE streaming, function calling and tool use, multimodal image input and structured outputs all behave exactly as they do against the official endpoints.
Which payment methods do you accept?
Alipay and WeChat Pay for users in mainland China, Stripe for international cards, and USDT for crypto. Every top-up appears under My Orders in the console.
Start now — you'll be integrated in minutes
Free credit on signup, no credit card, and nothing to cancel if you stop.