Text. Images. Video.
One gateway
for generative AI.

FastMetal is an OpenAI-compatible gateway to 303+ generative AI models. Point your existing code at one base URL, pay per token, and stop maintaining a different SDK for every provider.

Pay as you go·no subscription·no lock-in
303+ modelsOpenAI-compatiblePay-per-token35 providersCard or bank transferQualified invoices
POST api.fastmetal.ai/openrouter/videos/create200 OK
resp = requests.post(
  "https://api.fastmetal.ai/openrouter/videos/create",
  json={"model":"bytedance/seedance-2.0-fast",
        "prompt":"青白く発光する溶けた金属が六角形の鋳型に注がれ、火花がスローモーションで舞い散る…"},
)
Seedance 2.0 Fast · 8s 720p · ¥250 per video · real output
青白く発光する溶けた金属が六角形の鋳型に注がれ、火花がスローモーションで舞い散る…

Multi-provider AI is death by a thousand integrations.

Every provider ships its own SDK, its own auth, its own rate limits, its own billing, and its own idea of what a “message” is. Your agent’s reliability becomes a function of how many dashboards you can babysit at 3 a.m.

  • 01N providers = N API keys, N SDKs, and N monthly charges to reconcile.
  • 02A better model ships weekly — swapping it means a code change, not a config change.
  • 03One provider has an outage and your app goes down with it.

From many providers to one endpoint.

35 providers and 303+ models funnel through a single OpenAI-compatible URL. You integrate once and route to any of them by name.

303models
35providers
1endpoint
1payment

Three steps to your first call

From zero to a live request in about a minute. Keep the SDK you already use.

01

Create an account

Sign up and get an API key in under a minute.

02

Add credit by card or bank transfer

Card credits instantly. Bank transfer issues a qualified invoice, then credits on payment.

03

Change one base URL

Point your OpenAI client at FastMetal. Keep the rest of your code.

Everything you need, nothing you don’t

A gateway, not a framework. It gets out of the way of the code you already wrote.

Pay-per-token billing

Prepaid credits metered to the token. No seats, no subscriptions, no minimums — you pay for exactly what your app calls.

303+ models, one catalog

Every major provider behind a single key. Compare context length and arena rank before you commit.

anthropic/claude-fable-5#1
meta/muse-spark-1.1#10
google/gemini-3.1-pro-preview#15
qwen/qwen3.8-max#22

OpenAI-compatible API

Drop-in with the OpenAI SDK you already use. Chat completions, streaming, tools, and structured outputs — change the base URL, keep the rest.

The questions procurement asks, answered up front.

Using an overseas API directly means a dollar card statement and no invoice, which is where Japanese accounting stops the purchase. FastMetal is run by a Japanese company to Japanese billing requirements.

Billed in yen

Prices are shown and charged in Japanese yen: no exchange-rate drift in the monthly figure and no foreign-transaction card fees.

Qualified invoices (インボイス)

The operating company is a registered qualified-invoice issuer (T5010001015990); every credit purchase comes with one.

Bank transfer (prepaid)

Pay by card or by bank transfer. Choosing transfer issues a qualified invoice immediately, carrying a bank account dedicated to your organisation; credit is added automatically once the transfer is confirmed, with nothing to notify us about.

Prepaid, so spend is capped

The balance you buy is the ceiling. A bill above it cannot happen.

Models that run inside Japan

Models whose id starts with japan- run inference inside Japan. For work whose data cannot leave the country, they are one model name away on the same key.

Works from your agent, too

FastMetal ships a built-in MCP server. Add it to Claude Code, Codex, Cursor, or any MCP client with one command, and your agent gets inference, image, and video generation as tools — billed to the same prepaid key.

$ claude mcp add --transport http fastmetal \
    https://mcp.fastmetal.ai/mcp \
    --header "Authorization: Bearer ${FASTMETAL_API_KEY}"
askcomparegenerate_imagegenerate_videoanalyze_imagelist_models
See the MCP setup guide

Browse the catalog. Compare on the numbers.

303+ models ranked by arena performance, with transparent per-million-token pricing in your local currency.

Claude Fable 5anthropic/claude-fable-5
1M ctx¥1,787 in¥8,935 out / 1M
Text overall#1
Muse Spark 1.1meta/muse-spark-1.1
1.0M ctx
Text overall#10
Gemini 3.1 Pro Previewgoogle/gemini-3.1-pro-preview
1.0M ctx
Text overall#15
Qwen3.8 Maxqwen/qwen3.8-max
1M ctx¥357 in¥1,072 out / 1M
Text overall#22
GPT-5.5openai/gpt-5.5
1.1M ctx
Text overall#24
No sample models from that provider here — the full catalog has 303 across every provider.
Showing 5 of 303 modelsBrowse the full catalog
base_url = "https://api.openai.com/v1"
+base_url = "https://api.fastmetal.ai/v1"
Switching model = switching a string. Switching provider = switching a string. That’s the whole migration.

The things developers ask first

Is it really OpenAI-compatible?
Yes. Any OpenAI SDK or HTTP client works — set base_url to https://api.fastmetal.ai/v1 and use your FastMetal key. Chat completions, streaming, tool calls, and structured outputs are supported.
What does it cost?
You pay per token at each model’s published rate, drawn from prepaid credits. No subscription, no seat fees, no minimum spend.
Am I locked in?
No. It’s the standard OpenAI request shape — point your client back at any provider whenever you want. Your code doesn’t depend on us.
How long does integration take?
About a minute. Change one base URL, add your key, and your existing calls route through FastMetal.
Which providers are covered?
35 providers and counting — 303+ models in one catalog, growing as new models ship.
How does it compare with OpenRouter or going direct?
We keep honest comparison pages, including the cases where the other option fits better. vs OpenRouter · vs Anthropic direct · All comparisons

One key away from every model.

Create an account, add credits, change one line. Your agents get 303+ models behind one API key.

Prepaid credits·no subscription·cancel by simply not topping up