Aleph Alpha Pricing in 2026: Quoted Deals and Costs
Nov 10, 202614 min read
Aleph Alpha sells mostly through quoted deals, much as teams that wire up their own Activepieces workflows tend to negotiate custom terms for the pieces around them.
The German AI company behind the Luminous and Pharia language models had no public per-token rate for any model that I could confirm as of 10 November 2026.
To estimate a cost per run, you multiply tokens by a rate.
The rate is the part Aleph Alpha doesn't put on a page, because it sells to enterprises and public bodies through negotiated contracts rather than a self-serve rate card. Anyone who quotes you a precise monthly bill before talking to sales is guessing.
Aleph Alpha pricing today, plan by plan
The verified price table and check date
Read "quote only" as a hard stop: you can't budget the figure from public information. The table lists every Aleph Alpha model or plan I could name, with what Aleph Alpha publishes for each.
**Anyone who quotes you a precise monthly bill before talking to sales is guessing.
| Plan or model | Published price | Inclusions | Limits |
|---|---|---|---|
| Luminous Base | Quote only | Smallest Luminous text model | Quote only |
| Luminous Extended | Quote only | Mid-size Luminous text model | Quote only |
| Luminous Supreme | Quote only | Largest Luminous text model | Quote only |
| Luminous Supreme Control | Quote only | Supreme variant tuned to follow instructions | Quote only |
| Pharia-1 7B | Quote only | Smaller Pharia-family model | Quote only |
| On-prem licence | Quote only | Deployment inside your own infrastructure | Quote only |
You can't rank these options by cost before a conversation with sales, because every price cell reads the same way. The sizes are only a rough guide: larger models usually cost more per token, but Aleph Alpha hasn't published the multiples.
Aleph Alpha pricing details not published
You will not find per-token rates, any free tier, rate limits, or any batch or cached discount on public pages. Without them, you can't price even a small workload honestly from outside.
If you are running this arithmetic for your own team, see what the same workload costs on Activepieces.
Free tier, rate limits, and trial terms
Before you trust a monthly bill, get every gap below in writing, because Aleph Alpha publishes less on free usage, rate limits and discounts than a cost estimate needs.
As of 2026-11-10, I could not confirm any free allowance or its limit, so do not budget around one.
The Standard Pool is the default rate-limit pool, measured in tokens per minute, with a value I don't quote here. The Legacy Pool is a separate pool with its own tokens-per-minute ceiling, also unquoted here.
Assume no batch or cached-token discounts, because Aleph Alpha documents none. I have no confirmed credit amount or trial length, so my estimate assumes the first run costs full price.
Rate limits and concurrency
Before you size a batch job, ask which pool your model sits in, because that decides your ceiling. Limits are in tokens per minute, and they divide between a Standard Pool and a Legacy Pool. I have no confirmed concurrency cap.
When its tokens per minute can't absorb the volume, a nightly job that looks cheap on paper can stall, and the delay costs engineering hours instead of tokens.
If repeated prompts dominate your workload, this is the first line to raise in a quote, because a discount there would change the monthly total more than any other term in this section.
A worked monthly bill for one named workload
For 2,000 contracts a month, the bill is 16 million input tokens times the input rate, plus 1.6 million output tokens times the output rate. That assumes no cache discount. I have no confirmed rates, so the formula stands in for the dollar figure.
The workload and its token counts
On average, each contract runs to 8,000 input tokens and 800 output tokens. Divide the 16 million and 1.6 million monthly totals by 2,000 and you get those figures.
That's ten input tokens for every output token, so the input rate drives the bill and the output rate barely moves it.
Aleph Alpha monthly cost by pricing route
Read the last column first: a cost you can compute from a public page is a budget, and a cost that needs a quote is only a guess until a contract exists.
The table shows the formula for each route and whether you can check the rate without talking to sales.
| Route | Rates | Monthly cost | Published or quote |
|---|---|---|---|
| Luminous Base | Per-million input and output rates, to be confirmed in writing | 16 × input rate + 1.6 × output rate | Confirm with vendor |
| Supreme Control | Per-million input and output rates, to be confirmed in writing | 16 × input rate + 1.6 × output rate | Confirm with vendor |
| Mistral Small 4 | Per-million rates on Mistral's pricing page | 16 × input rate + 1.6 × output rate | Published |
| OpenAI GPT-6 Luna | Per-million rates on OpenAI's pricing page | 16 × input rate + 1.6 × output rate | Published |
| Claude Haiku 5.5 | Per-million rates on Anthropic's pricing page | 16 × input rate + 1.6 × output rate | Published |
| On-prem | Licence and hardware terms | Fixed fee, set by contract | Quote |
Fill in the three published rows with nothing more than a calculator. The on-prem row is a flat fee, so it doesn't scale with tokens at all.
At 10,000 contracts, the workload becomes 80 million input and 8 million output tokens, and every per-token row multiplies by exactly five.
The on-prem fee doesn't, so it may be the cheapest route at 5x and the most expensive at 1x. Get the break-even volume from the quote, not from this table.
Hidden Aleph Alpha costs after signing a contract
After signing, the costs that arrive are mostly outside any rate card: servers if you self-host, minimum commitments, paid support and onboarding, and model changes that force rework.
Aleph Alpha self-hosted infrastructure costs
Move to self-hosting and the meter shifts from tokens to hardware, power, and the people who keep it running.
A published study from KI-Beratung Unternehmen puts on-premise private AI for a 200-employee company at €118,600 in year one and €33,600 per year after that, which means first-year costs greatly exceed the ongoing annual expense.
For 500 employees it's €219,000 in year one and €54,000 afterwards.
The study is a general mid-market estimate, not an Aleph Alpha quote, so treat it as the shape of the curve rather than your number.
In year one, the 500-employee profile costs about 85% more than the 200-employee one (€219,000 vs €118,600), but only about 60% more thereafter (€54,000 vs €33,600), so the cost disadvantage of larger deployments narrows over time.
Minimum commitments, support and onboarding costs
If your usage comes in under the minimum term or spend that enterprise deals usually carry, you still pay it. Ask whether unused volume rolls over, and what happens at renewal if you renegotiate the price.
Expect vendors to price support tiers and onboarding as separate line items, while integration work with your own engineers or a consultancy sits outside the vendor's invoice entirely.
When a model is versioned, superseded or retired, each change can mean re-testing prompts, adjusting integrations, and re-validating outputs. Budget engineering hours for at least one migration during the contract, and read the vendor's changelog before signing, not after.
Before you sign, ask the sales team for answers to these in writing:
- What is the minimum commitment, and does unused volume carry over?
- Which support tier does the price include, and what do the others cost?
- Does onboarding cost extra?
- What hardware does self-hosting need, and who supplies it?
- What is the notice period before a model is retired?
Worth checking against a plan that does not meter every step: one credit covers a whole run on Activepieces.
Aleph Alpha pricing versus competitors
On price, Aleph Alpha is rarely the cheapest option. What its quotes buy is control over where models run, so get the written answers above before comparing it to anyone's rate card.
Per-token price and published rates
With no public rate card, there's no per-token number to put beside the others, so you can't budget from the website. That is the finding in the Aleph Alpha row. The table shows what each vendor publishes.
| Vendor | Input $/M tokens | Output $/M tokens | Public rate card | Free tier | Data residency or on-prem option | Check date |
|---|---|---|---|---|---|---|
| Aleph Alpha | Quoted per deal | Quoted per deal | No | Not verified | Deployment terms set in the contract | 2026-11-10 |
| Mistral | Varies by model | Varies by model | Yes | Not verified | Open-weight models such as Mistral Large 4 can be self-hosted | 2026-11-10 |
| OpenAI | Varies by model | Varies by model | Yes | Not verified | Not verified | 2026-11-10 |
| Anthropic | Varies by model | Varies by model | Yes | Not verified | Not verified | 2026-11-10 |
Per LLM Price, the tier bills run Mistral, OpenAI, Anthropic. At the small tier that's $12, $44 and $20 a month, so Mistral costs about a quarter of OpenAI.
At mid it's $80, $440 and $400, so Mistral is roughly a fifth of either rival. At flagship it's $112, $800 and $800, so OpenAI and Anthropic are identical and Mistral is about 86% cheaper. Aleph Alpha is absent because there's nothing to chart.
Sovereignty and deployment options for Aleph Alpha
As the tiers show, a public rate card can be far cheaper. A buyer who accepts an invoice several times larger is paying for something other than tokens: a say in where the model runs and who can see the data.
Mistral also publishes open weights, so "control" alone doesn't separate Aleph Alpha from it. The question for your contract is what Aleph Alpha guarantees in writing that a self-hosted open model wouldn't.
When a regulator, a customer contract or your own security policy dictates where inference happens, the premium pays off. If none of those apply, run your own workload through the Mistral, OpenAI and Anthropic figures first.
What Activepieces does about this
Activepieces can't publish Aleph Alpha's rate for you, and no tool can. What it can do is keep the part of the bill you control separate from the part a contract sets.
Activepieces runs your workflow on your own provider key, so whatever Aleph Alpha quotes you is billed on your Aleph Alpha account, not marked up through us.
When the quote arrives, you put its input and output rates into the 16 million and 1.6 million token formula above and get a monthly figure you can check against the invoice.
Our side of the bill is published. The Plus plan is $20 a month, and from that plan every AI step costs 1 credit, so the platform cost of the 2,000-contract workload doesn't depend on which model sits behind the step.
The model cost is the only number you're waiting on a quote for, and you can keep it apart from the rest.
For a vendor with no native connector, the generic HTTP step does the job described in the last answer: credentials in the secrets store, a request to the endpoint your contract names, and the text passed on to the next step.
The retry and per-run call cap that answer recommends are settings you add to that flow.
The same flow can later point at Mistral, OpenAI or Anthropic by swapping the endpoint and key, which makes the side-by-side comparison on your own workload a change to one step and not a rebuild.
The core is open source under the MIT licence, so if the contract you sign involves on-prem or private-cloud inference, the automation layer can run on your own infrastructure too. You can also reach your workflows from Claude, ChatGPT, Cursor or any MCP client.
MoneyGram and Moneypenny run Activepieces in production. That doesn't answer what Aleph Alpha will charge, but it means that once the quote exists, the rest of the cost is already a known number.
Frequently asked questions
Does Aleph Alpha have a free plan?
I have no verified free tier for Aleph Alpha's API to point to, so treat it as unconfirmed until the vendor's own pricing or sign-up page says otherwise. That means you shouldn't budget a prototype around free usage.
Ask for a time-limited trial or evaluation allowance in writing, and price the pilot as if it'll be billed.
Can I pay Aleph Alpha by credit card without a sales call?
I couldn't verify a public self-serve checkout, and the vendor's pricing is built around quoted deals. So plan on a conversation before any invoice exists.
The practical consequence: you can't compute a cost per run from a rate card, and any monthly figure you build is an estimate resting on your own assumptions.
If a self-serve path has appeared since this was checked, it'll be on the vendor's own site, and that page beats anything here.
Is Aleph Alpha cheaper than OpenAI?
Nobody can answer that without a quote, because one side of the comparison is a published per-token price and the other is a negotiated number.
Do the comparison on your own workload. Take your monthly volume of input and output tokens, price it at OpenAI's published rates, then ask Aleph Alpha to quote the same volume.
Compare the totals, and include any minimum commitment, setup fee or support charge, since a low unit price can still lose to a smaller bill with no floor.
Can I run Aleph Alpha models on my own servers?
Ask directly about on-premises or private-cloud hosting rather than assume it, since Aleph Alpha has positioned itself around sovereign and private deployment. Whether it is offered for the specific model you want, and on what licence terms, is a contract question.
Get the hardware requirements and the licence cost in the same quote, because running it yourself moves spend from per-token fees to GPUs and staff time.
How do I connect an Aleph Alpha model to my business tools?
Treat the model as an HTTP API call inside a workflow tool. The steps are the same for most vendors:
- Get API credentials under your contract and store them in the workflow tool's secrets store, not in the workflow itself.
- Add an HTTP request step that sends your prompt and receives the completion, following the vendor's API documentation for the exact endpoint and fields.
- Parse the response and pass only the text you need to the next step, such as writing a ticket comment or filling a CRM field.
- Add retries and a cap on calls per run, so a loop can't run up a bill you have no rate card to predict.
If your workflow platform has no native connector, the generic HTTP step does the same job.
Related reading
Running the numbers
See what the same workload costs here.
Free forever plan, and every paid plan self-hosts at no extra cost.
See pricing Talk to sales
