Drael API cost: US$ 70 to get in, reasoning you can't cap

The Drael API cannot be bought by the token. It comes only with a paid four-week or annual plan, so the smallest purchase is R$ 199 / US$ 70. Through a key every call reasons, the reasoning is billed as output, and your max_tokens is discarded. On NotHoly the entry is a US$ 5 top-up.

Data collected on 9 and 10 November 2026 from Drael's limits documentation, its chat documentation and from NotHoly pricing. We did not buy a Drael plan, so the numbers below are list prices and arithmetic.

#What the US$ 70 door costs per token

A weekly plan and a free account both get plan_required from the API, according to the quickstart. A project with light traffic therefore pays the full block whether it sends three million tokens or three hundred million.

Four weeks of use	Drael plan floor	Effective rate per million
3M tokens (2M in, 1M out)	US$ 70	US$ 23.33
12M tokens (10M in, 2M out)	US$ 70	US$ 5.83
181M tokens (151M in, 30M out)	US$ 70	US$ 0.39

The floor only matches Drael's own extra-usage meter near 181 million tokens every four weeks. Below that volume you are paying Drael for capacity you do not use. A metered provider with no minimum, like NotHoly, bills only for the tokens you send at the catalogue rate on pricing, so a CI job that reviews a handful of diffs a week, or a script for one bug bounty report, costs cents instead of the US$ 70 block.

The block does not renew automatically, and when it lapses the API stops. Extra usage bought in advance does not help either, because the terms say it "is spent while a paid plan is running". Pure prepaid does not exist on Drael.

#The meter, for comparison

Drael publishes one meter for extra usage and Enterprise.

Drael drael-v1 public meter, per million tokens, USD	Input	Cached input	Output
Extra usage and Enterprise	0.18	0.04	1.40

Output is the expensive line, and drael-v1 is an alias over weights Drael does not disclose, so you can read the price but not what you are buying. NotHoly meters per token as well, with each model's rate and its published base on pricing and models, so the number on the bill belongs to a model you can name.

#You cannot turn reasoning off or cap the output

Output is the expensive line of Drael's meter, and through an API key you do not control it. The chat documentation says "A key's turn always reasons", and that none, off and enable_thinking: false "are all answered at low". The limits page labels the output line "The answer, reasoning included".

The usual defence against a runaway answer is max_tokens. Drael drops it: "max_tokens is dropped on a turn that reasons", and "through a key the ceiling is always dropped". The documentation gives the reason. With reasoning on and max_tokens at 4000, the model "returned an empty answer in 22 of 24 runs". Drael removed the parameter and left the model's behaviour as it was.

For anyone budgeting an agent loop, this means the most expensive line has no per-call ceiling. The same page says the request field priority is discarded too.

#Four calls at once, and a rate nobody publishes

Every plan allows four answers in flight per account, counted across all keys. The fifth request gets a 429, and the limits page says "we do not retry it for you". To scale, the documentation recommends spreading work "across separate accounts", which means more subscriptions.

The per-minute rate is unpublished because it "is being set against real usage". Inside the plan, usage is "unlimited in volume within fair use", and the page does not say where fair use ends. You cannot plan throughput against a number that does not exist.

The documentation is frank about what sits behind the queue: "Everybody being served at once shares one machine, so a busy minute is slower for everyone in it." A plan's priority "buys your place in a line, never a faster answer". Drael publishes no capacity figure and no SLA. We read those sentences as a description of a small shared deployment, and we would not put a deadline on it.

#What else the docs leave to you

Limit on Drael	What the docs say	What it does to your costs
Context	128K on Pro, 248K on Max, "served no wider than the engine reports"	Long targets go in slices, and the window you paid for can be served smaller
Concurrency	4 per account, across all keys	Parallel agents fail with 429
max_tokens	Dropped on every call through a key	No per-call cap on output
Unknown model ID	Answered by drael-v1, no error	Config mistakes stay silent
API format	OpenAI-compatible only	Anthropic-format clients need an adapter
Hidden prompt	"we prepend a system prompt you never see"	You pay for behaviour you cannot inspect

NotHoly's API takes both OpenAI and Anthropic formats at api.notholy.com/v1, and each model's context window and per-token rate are on models and pricing. You send your own max_tokens and reasoning effort, and the gateway passes them through, trimming max_tokens only when your prepaid balance cannot cover the worst case. A large context window is something you pay for when you fill it, not a block you buy up front. The drael-v1 post covers the alias.

#Where each fits

Drael's API fits a project that pushes a large, steady volume through one account every four weeks and does not mind a subscription, forced reasoning or an output it cannot cap. NotHoly fits irregular or light use, a CI job, an agent loop you have to budget, or a test before you commit: prepaid credit, no plan to renew, and max_tokens and reasoning you set yourself. NotHoly has no Pix and runs no code on its servers, so if those are your requirement Drael has them and NotHoly does not.

#Frequently asked questions

#Does Drael have an API?

Yes. It is OpenAI-compatible and available only on paid four-week or annual plans.

#How much does the Drael API cost?

At least R$ 199 / US$ 70 for four weeks. The public meter for extra usage is US$ 0.18 input, US$ 0.04 cached input and US$ 1.40 output per million tokens.

#Can I use the Drael API on the weekly plan?

No. The weekly plan and the free account both answer plan_required.

#Can I disable reasoning on the Drael API?

No. Through a key the lowest setting is low, and max_tokens is discarded.

#What is the cheapest uncensored AI API?

For light or irregular use, one with no minimum. NotHoly starts at US$ 5. Venice AI vs NotHoly compares a third option model by model.

#Verdict

Drael's API is a subscription with an endpoint attached. You pay US$ 70 before the first token, lose the API when the block ends, and cannot cap or switch off the reasoning you are billed for. It makes financial sense only above roughly 180 million tokens every four weeks, pushed through four concurrent calls on a machine the docs say everyone shares. For anything smaller, add US$ 5 on NotHoly pricing, point an OpenAI or Anthropic client at api.notholy.com/v1 and read the bill after a week.