One API for DeepSeek, GLM, Claude and Grok

Model prices per 1M tokens

These are the rates a call debits on one-api.ai, in US dollars per million tokens, for prompts in the standard context tier. Each price is built from the Cloudflare list price named in the source column; the pricing page shows every step and the model comparison adds cached input, long-context tiers and other resellers.

Lookups cost $0.001 per call, which is one credit. Prices last reviewed 2026-10-31.

Use it with the OpenAI SDK

Keep the OpenAI client you already use and change the base URL. The SDK appends /chat/completions, so requests go to https://one-api.ai/v1/launch/chat/completions.

from openai import OpenAI
client = OpenAI(base_url="https://one-api.ai/v1/launch", api_key="YOUR_ONE_API_KEY")
print(client.chat.completions.create(model="@cf/deepseek-ai/deepseek-v4.1-flash-1017", messages=[{"role": "user", "content": "Hello"}]).choices[0].message.content)

The model field takes the provider model id shown on each model page. Streaming, max_tokens and the usual message roles work as they do with OpenAI. A remote MCP server at https://one-api.ai/mcp exposes the same models and lookups to agents.

How prices are set

Every model price starts from the provider's list price on Cloudflare. Cloudflare adds 5% when prepaid AI Gateway credits are bought and passes inference through at list, so the cost floor for us is the list price divided by 0.95. On top of that floor we add a posted margin: 4% for the two DeepSeek models, 15% for Claude Haiku 5.5 and Grok 4.7, and 50% for GLM 5.3. Nothing else is added and there is no subscription.

Calls are billed in credits. A call reserves its worst case before it runs, then settles to the usage the provider reports. Billing is exact to a millionth of a credit, so a call costs its tokens at the posted rate, even a very short one; each model page prices three example workloads.

one-api.ai, OpenRouter or the provider directly

OpenRouter offers hundreds of models from many providers, automatic fallback between hosts and a mature dashboard. If you need breadth or a model we do not list, it is the better choice. Check its own pricing page for current fees.

Calling a provider directly (DeepSeek, Zhipu, Anthropic or xAI) gets you list price with no reseller margin, every provider-specific feature and the provider's own support. It also means a separate account, key and invoice for each one.

one-api.ai is narrower: five models, four lookups that are not LLM calls at all, and one prepaid balance. Its point is that each price is posted with its source and date, and that the margin is stated rather than hidden. It is early access, run by a small team, without fallback routing.

Lookups on the same key

• Page audit API: The page audit API fetches one public URL and returns its on-page SEO signals as JSON, so a script, agent or spreadsheet can check titles and tags without running a crawler.
• Domain lookup API (RDAP + robots + sitemap): The domain lookup API combines an RDAP registration record with a quick look at the domain's robots.txt and sitemaps, returned as one JSON object.
• ads.txt and sellers.json checker API: The ads.txt checker API reads a host's /ads.txt and /sellers.json and returns both as structured JSON, which is easier to diff and validate than the raw files.
• Tranco rank checker API: The Tranco rank checker API looks a domain up in the latest Tranco top-1M list, a research ranking built to be harder to manipulate than single-source lists.

Each lookup is $0.001, with 25 free lookups per account each UTC month once billing opens.

Sample lookup

{
  "source": "one-api",
  "final_url": "https://example.com/",
  "status": 200,
  "title": "Example Domain",
  "meta_description": null,
  "canonical": null,
  "robots_meta": null,
  "headings": {
    "h1": [],
    "h2": [],
    "h3": []
  },
  "hreflang": [],
  "json_ld_types": [],
  "links": {
    "internal": 0,
    "external": 0
  },
  "truncated": false
}

Questions

What is one-api.ai?

An OpenAI-compatible API endpoint and four lookup endpoints behind one key, with every price published per 1M tokens or per call, and its source and date.

Is one-api.ai the open-source One API project?

No. one-api.ai is not affiliated with songquanpeng/one-api on GitHub or with Intel oneAPI. It is a hosted service operated by WakeUp Plus LLC.

Can I use the OpenAI SDK?

Yes. Point base_url at https://one-api.ai/v1/launch and pass your one-api.ai key as the API key. The chat completions call stays the same.

Can I sign up today?

Billing is not open yet. Early access is by request on the About page; prices on this site are what calls will debit once it opens.

How do I pay?

Once billing opens, by prepaid credit packs in US dollars. Calls debit credits; there is no subscription.