Pay per answer, not per month.
Evanesca has no subscription. You add credit from €2 when you choose, VAT included, with an invoice for every purchase; credit never expires. Each answer, and each step of an Agent task, is priced by the tokens it uses, at its model's rate, VAT included, and its exact cost appears right under it.
Prices checked on 20 Oct 2026
Opening soon: what follows describes the service as it will work.
- TO START
- €2 of creditNo subscription
- YOUR CREDIT
- never expiresAn invoice each time you add credit
- Subscription
- Noneno monthly fee, no seats
- Lowest price
- €0.168Gemma 4 26B, Germany (AKI.IO), per million input tokens
How billing works
- Sign in, add credit from €2Sign in with your email, then add credit to your account from €2, VAT included: an invoice is e-mailed for every credit purchase.
- Ask, or hand over a taskEach answer — and each step of an Agent task — is priced by its tokens: what you send (input) at the model's input rate, what it writes (output) at its output rate. Tokens a model spends reasoning count as output.
- Read the cost under the answerThe exact cost appears under each answer — the tokens the operator reports, times the prices on this page — and comes off your credit.
- Add credit when you need toAny time, from €2, VAT included. Credit never expires, and there is no subscription to cancel.
Every model's price
Per million tokens, VAT included. “Quick question”: 300 tokens in and 400 out. The 17 models Evanesca serves all run in Europe — in France, in Mistral AI's EU region, in Germany or in Switzerland: the “Where” column gives each one's place. Your messages never leave Europe.
| Model | Run by | Where | Input / M tokens | Output / M tokens | Quick question |
|---|---|---|---|---|---|
 Mistral Large 4 | Mistral AI | EU region of Mistral AI | €2.5133 | €7.7246 | €0.0038 |
 DeepSeek V4.1 Flash | AKI.IO | Germany (AKI.IO) | €1.008 | €3.024 | €0.0015 |
 GLM-5.3 | AKI.IO | Germany (AKI.IO) | €3.024 | €9.24 | €0.0046 |
 Qwen3.8 27B | Scaleway | Paris, France | €1.008 | €5.544 | €0.0025 |
| DeepSeek V4 Flash | Scaleway | Paris, France | €0.672 | €1.344 | €0.0007 |
| Apertus v1.5 70B | Infomaniak | Switzerland (Infomaniak) | €1.2936 | €4.62 | €0.0022 |
| GLM-5.2 | Scaleway | Paris, France | €3.024 | €9.24 | €0.0046 |
| Qwen3.6 35B | Scaleway | Paris, France | €0.42 | €2.52 | €0.0011 |
 Gemma 4 26B | AKI.IO | Germany (AKI.IO) | €0.168 | €0.84 | €0.0004 |
| Mistral Small 4 | Mistral AI | EU region of Mistral AI | €0.2772 | €1.1088 | €0.0005 |
| Qwen3.5 397B | Scaleway | Paris, France | €1.008 | €6.048 | €0.0027 |
 GPT-OSS 120B | Scaleway | Paris, France | €0.252 | €1.008 | €0.0005 |
| Qwen3 Coder 30B | Scaleway | Paris, France | €0.336 | €1.344 | €0.0006 |
| Qwen3 235B | Scaleway | Paris, France | €1.26 | €3.78 | €0.0019 |
| Mistral Small 3.2 | Scaleway | Paris, France | €0.252 | €0.588 | €0.0003 |
 Llama 3.3 70B | Scaleway | Paris, France | €1.512 | €1.512 | €0.0011 |
| Pixtral 12B | Scaleway | Paris, France | €0.336 | €0.336 | €0.0002 |
Llama 3.3 is licensed under the Llama 3.3 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. Built with Llama. Compare the models side by side
What the chat's features cost
The same rules as for messages: VAT included, no subscription, and the exact cost under the answer or the task.
| Feature | How it is billed | Price |
|---|---|---|
| Agent mode | Each step is billed like a message, at the model's price; web calls and images on top. You set a cap for each task — €0.50, €1, €2, €5 or €10 — never above your balance: everything counts toward it. | The model's price, capped at €2 a task by default |
| Web search | Off by default, in the chat and in Agent mode. Per call, through Linkup, a French company: each search and each page read, plus the tokens the model spends reading the results (8 at most). Each round of searches is one more call to the model, and while search is on, the description of its two tools adds a few hundred input tokens to each message. | €0.0076 a search · €0.0015 a page read |
| Image generation | By the second of generation, measured around the call and rounded up to the next second. | €0.00924 a second — about €0.02 a FLUX.1 [schnell] image, €0.05 a PhotoMaker V2 portrait |
| Dictation | By the second of audio, 60 s at most per dictation. Transcribed by Voxtral Mini at Mistral AI, a French company. Mistral AI keeps API inputs and outputs “for thirty (30) rolling days to monitor abuse” (Mistral's privacy policy). Mistral AI lets its customers object to training on that data, and Evanesca's account objects. | €0.000084 a second — about €0.005 a minute |
| Memory | Off by default. When on, a small call to Mistral Small 3.2 (Scaleway, Paris, France) after each answer updates your profile, billed like a message; the profile — 4,000 characters at most, kept in your browser — goes with each message as input tokens. | Billed like a message |
| Titles and follow-up questions | After each answer, a small call to Mistral Small 3.2 (Scaleway, Paris, France) suggests follow-up questions; a new conversation also gets a title, and a long one a summary. Billed like a message; the cost shows in your account's usage history. | Billed like a message |
| Read-aloud | Your browser's own local voices: the text never leaves your device. | Free |
| Auto mode | Evanesca picks the model from signals in your message — an image, code, a long text, deep reasoning — among its models, all run in Europe. No extra call, no routing tokens. | No routing fee: the chosen model's price |
Embeddings, for search and RAG
An embedding model turns text into a vector for semantic search. It writes nothing: only input tokens are billed. Each one runs in France, at Scaleway.
| Model | Run by | Where | Dimensions | Context | Input / M tokens |
|---|---|---|---|---|---|
| Qwen3 Embedding 8B Qwen · Apache-2.0 | Scaleway | Paris, France | 4,096 | 40,960 tokens | €0.168 |
| BGE Multilingual Gemma2 BAAI · Gemma Terms of Use | Scaleway | Paris, France | 3,584 | 8,192 tokens | €0.168 |
BGE Multilingual Gemma2 falls under the Gemma Terms of Use: Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms. Using embeddings
Worked examples
The four answers of the sample receipt on the home page, at each model's real rate.
| Answer | Model | Tokens in | Tokens out | Cost |
|---|---|---|---|---|
| Contract review | Mistral Small 3.2 | 1,750 | 430 | €0.0007 |
| Q3 regional analysis 3 files | Qwen3.5 397B | 5,700 | 200 | €0.0070 |
| AI Act question | GLM-5.2 | 300 | 400 | €0.0046 |
| Whiteboard to tasks 1 image | Qwen3.6 35B | 1,500 | 300 | €0.0014 |
| Total | €1.99 left of €2 of credit | €0.0136 | ||
How the first line adds up: 1,750 input tokens × €0.252 per million = €0.000441, plus 430 output tokens × €0.588 per million = €0.000253, so €0.000694 — €0.0007 once rounded in the table.
For teams
A Business workspace adds what a company asks for: a signed Art. 28 DPA, SSO and roles, VAT invoices, spending limits — on top of the cap of each Agent task — and passport exports. Evanesca for teams
Frequently asked questions
Is there a subscription?
No. There is no monthly fee and no seat to pay for: you pay for each answer by the tokens it uses, from prepaid credit.
What does €2 of credit cover?
About 8,500 quick questions on Pixtral 12B, the cheapest model, or about 430 on GLM-5.2, the most expensive — counting 300 tokens in and 400 out per question. In the chat, small automatic calls (a new chat's title, the suggested follow-ups) are billed too: on the cheapest model they can cost about as much as the question itself.
Does my credit expire?
No, never — not even if you close your account: the credit stays attached to it, and you find it again if you come back. Closing the account erases its personal data, except what the law requires us to keep and what is needed to give you your credit back: the account's identifier and its balance.
What is the smallest amount of credit I can buy?
€2, VAT included. You choose when to add credit: nothing is ever charged automatically.
Do I get an invoice?
Yes: an invoice is e-mailed automatically with every credit purchase, as a PDF showing the VAT.
Is VAT included?
Yes. Every price on this site includes VAT: the per-token rates, the credit amounts and the cost shown under each answer.
How do I know what an answer cost?
The exact cost appears under each answer: the tokens the operator reports, multiplied by the prices on this page.
Are reasoning tokens billed?
Yes, as output, at the model's output rate: a model that thinks before answering writes more tokens than it shows. They are part of the cost shown under the answer.
What do web search, images and Agent mode cost?
Web search, off by default, is billed per call: €0.0076 a search and €0.0015 a page read, plus the tokens the model spends reading the results. Images are billed by the second — about €0.02 each. An Agent task is billed step by step at the model's price, web calls and images included, within the cap you set: €2 by default.
How do I pay for credit?
Online, from your account: by card or SEPA bank transfer, through Stripe. An invoice comes with every credit purchase.