API Pricing

Pay for the tokens you use. No monthly minimum, no per-seat fee, and no tier that unlocks features — every key gets everything.

$5.00

free credit to start

monthly minimum

per token

billed as each call finishes

Rates

Model	Size	Input / 1M	Output / 1M	Context
SoreQen S1 Mini `soreqen-s1-mini`	0.8B	$0.05	$0.15	256k
SoreQen S1 `soreqen-s1`	2B	$0.10	$0.30	256k
SoreQen S1 Mega `soreqen-s1-mega`	4B	$0.25	$0.75	256k

Prices are in US dollars, per million tokens, and are the same whatever your chat plan is. There is no volume tier and no discount to negotiate at this scale — the number above is the number.

Reasoning tokens are billed as output. They are generated tokens: asking for more thinking is asking the GPU for more work. This is the single largest thing that moves a bill, and it is under your control on every request.

What a real call costs

Rates per million are hard to hold in your head. These are ordinary requests, priced from the table above.

Request	Tokens	S1 Mini	S1	S1 Mega
A short question A sentence in, a paragraph out, no reasoning.	60 → 250	<$0.0001	<$0.0001	$0.0002
A support reply A ticket plus a page of context, a considered answer with light reasoning.	1,500 → 1,200	$0.0003	$0.0005	$0.0013
Extraction from a document Ten pages in, a small JSON object out, structured output with no thinking.	12,000 → 300	$0.0006	$0.0013	$0.0032
A hard reasoning task A problem worth High effort — most of the cost is the thinking.	800 → 6,000	$0.0009	$0.0019	$0.0047

Put another way: $5.00 of starting credit is roughly 9,803 support replies on SoreQen S1, or 121,951 short questions on SoreQen S1 Mini. Enough to build something and find out whether it works.

How reasoning effort changes the bill

Effort decides how many tokens the model spends thinking before it answers. The prose it finally writes barely moves; the thinking is what you are paying for.

Setting	Thinking budget	A 1k-token question on SoreQen S1
`thinking: false`	none	$0.0002
`effort: "low"`	0 tokens	$0.0002
`effort: "medium"`	512 tokens	$0.0004
`effort: "high"`	2,048 tokens	$0.0008
`effort: "max"`	16,384 tokens	$0.0051

The budget is a ceiling, not a quota — an easy question finishes early and is billed for what it actually used. But budget the worst case when you are sizing a workload, because a hard question will use it.

Concurrency, limits and capacity

There is no per-key concurrency cap and nothing to buy to raise one. Requests queue per model and are served by the GPU fleet behind it; under load a request waits rather than being rejected.

Concurrent requests	Unmetered. Shared fleet capacity, served fairly.
Requests per minute	No fixed limit. Sustained load is capacity, not policy.
Context window	262,144 tokens entitled; GET /v1/models reports what is served today.
Output per reply	32,768 tokens, thinking included.
Tools per request	128 function definitions.
Images per request	4.
Request timeout	15 minutes.
Active keys per account	25
Spend cap	Per key, per calendar month, set by you. Optional but recommended.

Cold starts. The fleet scales to zero when nobody is calling, so the first request after a quiet period starts a GPU and can take a minute or more. It answers 503 worker_cold_start rather than hanging, and the retry is fast. Keeping capacity permanently warm costs real money per hour, so it is not the default — if your workload needs it, ask and it can be arranged.

How billing works

Prepaid, not invoiced. You hold a balance and calls draw it down as each one finishes. There is no bill at the end of the month and no way to run up a debt.

Charged on completion, including cancellations. A stream you disconnect from is billed for what was generated before you left — the GPU did that work. Otherwise cancelling repeatedly would be a way to use the service free.

Refused before it runs, not overdrawn after. A request whose prompt alone costs more than your balance is refused with 402, and a reply is capped at what the remaining balance can pay for. A call cannot leave an account overdrawn by more than a rounding error.

Every call is itemised. Model, tokens in and out, latency and cost, per request, in the console. Nothing is aggregated away.

This is not the chat subscription

The two are separate products that share a sign-in, and neither gates the other. A free account with credit reaches every model through the API; Pro grants no API credit at all.

	Chat plans	API
You pay	A monthly subscription	For the tokens you use
You get	An allowance that refills	A balance that does not
Models	Depends on the plan	All three, always
Where	soreqen.com	Your own code

Chat plans →·API documentation →

Every account starts with $5.00 of credit. Create a key and spend some of it.