Frontier models, a fraction of the price.
A drop-in gateway for Claude and GPT that speaks both the Anthropic and OpenAI APIs. Point Claude Code, Cline, or any SDK at LajuAPI and keep working — prepaid, per-token, priced in USD. No subscription, no card on file.
Accounts are invite-only while billing is being finalised.
Two lines of config.
Change the base URL, keep your code. Streaming, tool use, and prompt caching pass through unchanged.
# point Claude Code at LajuAPI
export ANTHROPIC_BASE_URL=https://api.lajuapi.com
export ANTHROPIC_AUTH_TOKEN=sk-laju-••••••••
claudefrom openai import OpenAI
client = OpenAI(
base_url="https://api.lajuapi.com/v1",
api_key="sk-laju-••••••••",
)Any client that lets you set a custom base URL works.
Pay per token, in USD.
Prepaid balance drawn down per request. No subscription, no auto-renew, no card kept on file. Prices are per 1M tokens.
| Model | Input | Output | List price | Delta |
|---|---|---|---|---|
| Anthropic · Claude | ||||
| claude-opus-4-8 | $0.55 | $2.75 | $5.00 / $25.00 | −89% |
| claude-opus-5 | $0.55 | $2.75 | $5.00 / $25.00 | −89% |
| claude-sonnet-5 | $0.25 | $1.25 | $3.00 / $15.00 | −92% |
| claude-sonnet-4-6 | $0.35 | $1.75 | $3.00 / $15.00 | −88% |
| claude-haiku-4-5 | $0.12 | $0.60 | $1.00 / $5.00 | −88% |
| OpenAI · GPT | ||||
| gpt-5.6-sol | $0.22 | $1.32 | $5.00 / $30.00 | −96% |
| gpt-5.6-terra | $0.11 | $0.66 | $2.50 / $15.00 | −96% |
| gpt-5.6-luna | $0.05 | $0.30 | $1.00 / $6.00 | −95% |
Every model is served under its own name — Claude requests reach Claude, GPT requests reach GPT, and nothing is relabelled in between. Balance never expires. Top up with USDT, or — if you're in Indonesia — QRIS, e-wallet, or local bank transfer. During the beta, top-ups are issued manually as redemption codes. Indicative beta pricing, subject to change; balance you already bought is honoured at the rate you paid. “List price” is each model provider's own published rate as of July 2026, shown for comparison only.
The honest version.
Cheap access is easy to promise and easy to get wrong. Here is what we commit to — and what we don't.
Best-effort, no SLA
We run multiple upstream suppliers and fail over between them. We don't sell an uptime guarantee we can't back — we publish a real status page instead, and it's wired into the bar at the top of this page.
No silent model substitution
You get the model you asked for. We never quietly serve a cheaper model behind a more expensive name — we run automated checks against our own suppliers for exactly this.
Prepaid only
No subscriptions, no stored cards, no surprise charges. You top up what you want to spend; when it runs out, requests stop.
Predictable, typed errors
Every failure returns a documented error with a request ID — never a raw upstream message, and never a 401 that wrongly blames your key when the fault is ours.
Honest about logging
We keep request metadata for billing and abuse handling, and prompt bodies only briefly for debugging. Requests are processed by third-party model infrastructure whose logging we don't control — so we don't claim “zero logging”.
Independent service
LajuAPI is not affiliated with, endorsed by, or a reseller acting on behalf of any model provider. We're compatible with their API formats — that's all.
Questions worth asking.
Is this an official Anthropic or OpenAI service?
No. LajuAPI is an independent gateway that speaks the same API formats, so existing SDKs and coding tools work unchanged. We have no affiliation with or endorsement from any model provider.
How can it be this much cheaper?
We buy capacity in bulk through the aggregator market and resell it at a margin, on our own infrastructure and our own keys. That's the whole trick — there's no special deal being implied and no partnership being claimed. It also means our pricing tracks what that market does, which is why we publish “indicative” rates and honour whatever rate you already paid.
How do I pay?
Prepaid balance in USD. Internationally that's USDT; in Indonesia you can also use QRIS, e-wallet, or bank transfer. During the beta you pay and we issue a redemption code you apply to your balance — automated top-ups land with general availability.
What happens when an upstream model goes down?
Requests fail over across our suppliers first, so most upstream hiccups never reach you. If every route is unavailable you get a typed, retryable error in the standard format for the endpoint you called, with a request ID you can send us. We never forward raw upstream errors.
Do streaming and prompt caching work?
Yes. Streaming responses, tool use, and prompt caching pass through — which is what makes agentic tools like Claude Code usable and cheap on long sessions.
Are my prompts stored?
We retain request metadata — timestamps, model, token counts, latency, request ID — for billing and abuse handling. Prompt and completion bodies are kept only for a short debugging window. Your requests are processed by third-party model infrastructure, and we're transparent that we don't control their retention policies.
Can I get an account now?
Accounts are invite-only while the billing flow is being finalised. The API, dashboard, and status page are already live — public signup opens together with automated top-ups.
Point your tools at one endpoint.
The API, dashboard, and status page are live. Signup opens with automated billing.