$ curl llmtap.vercel.app --同一模型 --便宜10倍
在 gpt-5.6 系列上，相比 OpenAI 官方价格，你每个 token 少付这么多。同样的前沿模型（GPT、Claude、Gemini），兼容的 API，按精确 token 计费。无订阅，无套餐速率限制。
开始即送 $50 免费用量。无需银行卡。
$ curl https://llmtap.dev/api/v1/chat/completions \
-H "Authorization: Bearer sk-live-your-key" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5.6-sol",
"messages": [{"role": "user", "content": "hello"}]
}'30 秒创建免费账户。无需银行卡。
在仪表板中生成带名称的密钥。请立即复制：只显示一次。
把任何兼容 OpenAI 的客户端指向 https://llmtap.dev/api/v1。一行代码，仅此而已。
同样的 SDK、同样的代码、同样的模型，只需一小部分成本。
Sign up with your email. Top up with USDT or USDC (BSC network) and your balance is credited automatically once the network confirms. Promotional redeem codes are also supported.
From the dashboard you create named keys, revoke them anytime, and see per-request usage with exact tokens.
The API is compatible with the OpenAI format: point base_url to llmtap, use your key, and your code stays the same.
Exact billing based on tokens reported by the upstream. Official is the list price of the corresponding provider; savings are computed on input.
| 模型 | 提供商 | input | output | 官方 in/out | 节省 |
|---|---|---|---|---|---|
| gpt-5.6-sol | OpenAI | $0.50 | $3.00 | $5.00/$30.00 | −90% |
| gpt-5.6-terra | OpenAI | $0.25 | $1.50 | $2.50/$15.00 | −90% |
| gpt-5.5 | OpenAI | $0.50 | $3.00 | $5.00/$30.00 | −90% |
| gpt-5.4 | OpenAI | $0.25 | $1.50 | $2.50/$15.00 | −90% |
| codex-auto-review | OpenAI | $0.25 | $1.50 | $2.50/$15.00 | −90% |
| claude-opus-4.8 | Anthropic | $0.80 | $4.00 | $5.00/$25.00 | −84% |
| claude-sonnet-5 | Anthropic | $0.32 | $1.60 | $2.00/$10.00 | −84% |
| claude-haiku-4.5 | Anthropic | $0.08 | $0.32 | $0.50/$2.50 | −84% |
| claude-fable-5 | Anthropic | $8.00 | $40.00 | $10.00/$50.00 | −20% |
| gemini-3.1-pro | $0.53 | $3.20 | $2.00/$12.00 | −73% | |
| gemini-3.5-flash | $0.40 | $2.40 | $1.50/$9.00 | −73% |
按你的月用量，把我们的转售价格与官方价格对比。
按 input 与 output 10:1 的比例计算，即 Cursor 或 Claude Code 的典型使用模式。
任何兼容 OpenAI 的客户端。改一次 base URL，其余保持不变。
base_url = "https://llmtap.dev/v1" api_key = "sk-..."
If your code already talks to OpenAI, pointing it to llmtap means changing base_url and the key. Streaming, function calling and the error format behave the same.
And if the upstream fails, the charge is refunded automatically. You only pay for completed responses.
from openai import OpenAI
client = OpenAI(
api_key="sk-live-your-key",
base_url="https://llmtap.dev/api/v1",
)
res = client.chat.completions.create(
model="claude-sonnet-5",
messages=[{"role": "user", "content": "hello"}],
)
print(res.choices[0].message.content)We buy capacity in bulk through subscription pools and route your requests to the official servers. Same model, shared cost. It is the same mechanism API proxies in Asia have used for years.
The request goes to the official upstream unmodified. You can run your own benchmarks against the official API and compare: same weights, same results. Plus every request is logged with exact tokens in your dashboard.
You don't pay. We lock an estimate before calling and settle against the real usage reported by the upstream. If the stream breaks or the provider errors, the hold is fully refunded to your balance automatically.
With crypto: USDT or USDC on the BSC network, processed by NowPayments. Balance is credited only once the transaction confirms on-chain. No cards, no monthly subscription: you pay per token consumed.
No. You change base_url and the API key. It works with the OpenAI SDK, with curl, and with any client that speaks the /v1/chat/completions format.
Real. Requests go to the official upstream unmodified. You can benchmark the same model against the official API and compare results: same weights, same output.
We buy capacity in bulk through subscription pools and pass the savings to you. Same mechanism API proxies have used for years. No retail margin, no per-seat licensing overhead.
Yes. We route through redundant upstream pools. If one provider degrades, requests fail over to the next available. Typical latency is within 5% of the official API. No request storage, metadata-only logging.
Requests are not stored or logged beyond metadata (model, tokens, timestamp) needed for billing. Prompt and response content is not retained. No training, no third-party sharing.
In Cursor: Settings → Models → Add OpenAI-compatible provider → base URL https://llmtap.dev/api/v1 + your key. In Claude Code: set ANTHROPIC_BASE_URL to https://llmtap.dev/api/v1 and use your key as the API token. Works with any OpenAI-compatible client.
We are pay-as-you-go only for now: you pay per exact token consumed, nothing more. Subscriptions with monthly caps are on the roadmap.