OpenRouter list prices · own capacity where we can

Tokens produced here.
Priced like OpenRouter.

One OpenAI-compatible endpoint for DeepSeek, Qwen, GLM, Claude and GPT. Same published rates, every token metered per request.

See the rate card How tokens are made

Own GPUs for open-weight models Per-request ledger Zero retention

Rate card

Per million tokens, at OpenRouter prices

No subscription, no minimum, no credits that expire. If the published rate moves, yours moves with it.

USD per 1M tokens · updated against OpenRouter All models →

Model	Context	Input	Cached in	Output	Served by

List prices tracked from OpenRouter's public rate card. Open-weight models run on our Strix Halo nodes; Claude and GPT are routed to their provider at the same list price — never silently swapped for a cheaper model.

Token factory

Where the tokens come from

Prompts land in our gateway, get placed on hardware we own, and leave as metered tokens.

Own-weight models: produced on our silicon. Closed models: routed, still metered by us.

Token Visor

Every token is a row

Not a monthly aggregate you have to trust — a per-request ledger you can export.

tokens produced

requests metered

0 ms

median TTFT

metered today

usage_ledger · last hour reported_final

Model	Prompt	Completion	Cached	TTFT	Cost

Quickstart

Three lines to switch

Same SDK. Base URL, key, model name — that is the whole migration.

Sign in

Invite-only: your invitation creates the account and your personal tenant.

Swap the base URL

Streaming, tools and JSON mode behave exactly as before.

Watch the ledger

Every call lands with its served model and cost.

client.py

# before
client = OpenAI(base_url="https://openrouter.ai/api/v1",
                api_key=os.environ["OR_KEY"])

# after — same price, tokens metered by us
client = OpenAI(base_url="https://api.sealioncloud.ai/v1",
                api_key=os.environ["SEALION_KEY"])

r = client.chat.completions.create(
  model="deepseek/deepseek-v4.1-flash",
  messages=[{"role": "user", "content": "…"}])

# ledger served_model deepseek-v4.1-flash
# prompt 18,204 · completion 612 · cached 4,096
# cost $0.00281

FAQ

The four that matter

Why are your prices the same as OpenRouter?

Because a rate card nobody can compare is a rate card nobody trusts. We track their published prices, and where we run the weights ourselves our cost is low enough to hold that price.

Which models do you actually produce?

Open-weight models — DeepSeek, Qwen, GLM — run on our Strix Halo nodes. Claude and GPT are routed to their provider and charged at list price. The ledger always names the model that ran.

What if OpenRouter changes a price?

Ours follows on the same revision. The rate applied to each request is stored with the row, so a historic invoice never rewrites itself.

What happens to my prompts?

Processed in transit, never stored for training, never resold.

Start on the cheapest model. Stay for the ledger.

DeepSeek v4.1 flash at $0.045 per million input tokens.

Tokens produced here.Priced like OpenRouter.