clinics, shops, training centres
Answering customers around the clock
Opening hours, bookings, sizes, order status. Clinics, shops and training centres answer on the spot, at night and in August.
500 answers a day≈ 1.43 € / month
Token Factory, LLM inference API
Token Factory: the OpenAI-compatible LLM API
Add AI to your website, shop or app without running a single server. Your product sends a question, our models answer it, and you pay only for the text processed. It works with the OpenAI and Anthropic SDKs you already use. Change the base URL and the key, and you are migrated.
No code needed to get started: activate your welcome credit with a card and the panel gives you your key and ready-to-copy examples.
how it works
What Token Factory is, in one minute
A large language model (LLM) is the kind of AI behind ChatGPT: it reads text and writes text back. Token Factory lets your own software use models like that through an API, which is simply a web address your app sends requests to. You do not install anything, buy hardware or manage updates. We take care of the machines and the models; you send the question and get the answer. The models are open source, so you are not tied to a single vendor, and the price is public and the same for everyone. A “token” is the unit the models count in: a million tokens is roughly 750,000 words.
- 01
Your app sends the request
From your website, your shop, your CRM or your internal tool. Your customers see nothing different.
- 02
Our models process it
On NVIDIA's current-generation chips. Your data is never used to train any model.
NVIDIA GPU - 03
The answer comes straight back
Written in the tone you set, at 3 a.m. and in August too. Each answer costs a fraction of a cent.
Wiring it into your site or app takes a developer one afternoon. If you have no developer, write to us and we will walk you through it.
what people use it for
What businesses use it for, and what it costs them
Three everyday jobs, with the real cost right beside each one.
accountants, back office
Reading invoices and paperwork
Hand it the photo or the scan and get the data back in order: dates, amounts, who invoiced you.
1,000 invoices scanned≈ 0.49 €
marketing, retail
Writing in your own voice
Product descriptions, emails and posts that sound like your brand.
300 product descriptions≈ 0.03 €
the brains on offer
Pricing per million tokens
One rate for everyone, no volume tiers, VAT included. “Input” is what you send; “output” is what the model writes back. You load credit in euros and it goes down as you use it.
| model | input €/Mtok | output €/Mtok | details | |
|---|---|---|---|---|
GLM 5.3 NVFP4 An all-rounder for writing and coding, with tool use. ChatCodingReasoningTools | ChatCodingReasoningTools | 1.206 € | 3.789 € | details |
Qwen3.8 27B Strong at coding and at chaining tasks on its own. ReasoningJSON outputTools | ReasoningJSON outputTools | 0.10 € | 0.30 € | details |
Qwen3 14B Thinks before answering, and you decide how much: a balance of quality and spend. ChatReasoningTools | ChatReasoningTools | 0.20 € | 1.00 € | details |
Qwen3.5 9B Light and very fast, for simple tasks at high volume. ChatReasoningTools | ChatReasoningTools | 0.10 € | 0.15 € | details |
ALIA 40B Instruct Spain's public model, trained in Spanish, Catalan, Basque and Galician. Chat | Chat | 0.20 € | 2.00 € | details |
Gemma 4 26B A4B Understands text and images at once without being a heavy model. ReasoningToolsVision | ReasoningToolsVision | 0.12 € | 0.30 € | details |
Nemotron Nano 12B VL Reads scanned documents and photos: invoices, contracts, forms. VisionOCR | VisionOCR | 0.06 € | 0.24 € | details |
Qwen3-VL 32B OCR Document reading for awkward layouts: tables, columns and handwriting. VisionOCR | VisionOCR | 0.10 € | 0.35 € | details |
Nemotron 3.5 Lightning 30B Reasons and uses tools while spending very little, built to answer fast. ReasoningTools | ReasoningTools | 0.12 € | 0.30 € | details |
Whisper large-v3 Turns audio into text: meetings, calls, interviews and videos. Transcription | Transcription | 0.0015 €per minute of audio | details | |
Qwen3-TTS CustomVoice Reads your text aloud in the voice you choose. TTSPreset voices | TTSPreset voices | 19.00 €per million characters | details | |
Qwen3-TTS Base (clone) Speaks in the voice you want: give it a few seconds of sample audio and it imitates it. TTSVoice cloning | TTSVoice cloning | 5.00 €per million characters | details | |
ACE-Step Music Composes a complete song from lyrics and a style. Text to musicCover versionsRedo sectionsSplit stemsEdit the song | Text to musicCover versionsRedo sectionsSplit stemsEdit the song | 0.50 €per song | details | |
The price is per million tokens: roughly 750,000 words.
catalogue
13 open-source models, one API key
Every model in the catalogue is open source. You switch between them by changing one word in the request.
Qwen API: Qwen3.8 27B, Qwen3 14B, Qwen3.5 9B and Qwen3-VL 32B OCR
Alibaba's open models, from the lightest to the strongest at coding. The OCR version reads scanned documents.
GLM API: GLM 5.3 NVFP4
The largest model in the catalogue (753B MoE), with a context of up to one million tokens, for the hardest writing and coding jobs. Above 256,000 tokens per request, write to us and we will raise it.
Gemma API and Nemotron
Google's Gemma 4 for text and images; NVIDIA's Nemotron family for low-cost reasoning and document reading.
ALIA 40B: the Spanish public model
Trained by the Barcelona Supercomputing Center in Spanish, Catalan, Basque and Galician. For products that have to speak Spain's languages well.
Whisper API and text-to-speech API
Whisper large-v3 turns audio into text; Qwen3-TTS turns text into natural voices, with preset voices or by cloning one. ACE-Step composes music from lyrics.
what it costs
Like a prepaid card: you top up and you use it.
You load credit in euros and each answer takes its share. No monthly fees, no lock-in and no nasty surprises: if the credit runs out, it simply stops.
- You start with €5 on the house: enough for around 50,000 answers.
- You see the daily spend in your panel, in euros with VAT.
- If you like, your Sandbox hours can come out of this credit too.
get a feel for it
It would cost you1.43 €a month
A shop with real traffic. Less than a coffee a month, answering at any hour. Worked out with Qwen3.5 9B, the cheapest model in the catalogue, assuming 500 input tokens and 300 output tokens per answer. A more powerful model costs more, and you can switch whenever you like.
for developers
Drop-in replacement for the OpenAI and Anthropic SDKs
If your code already talks to OpenAI or Anthropic, it already talks to Token Factory. Point the SDK at our base URL, paste your key, pick a model. Messages, streaming, function calling, JSON mode and reasoning work exactly as before.
Supported: /v1/chat/completions, /v1/messages (Anthropic format), /v1/audio, streaming, tool calling, JSON output, reasoning. Also from the terminal with the GPU Flow CLI, and from Claude Desktop or Cursor through our MCP server.
privacy
GDPR, certifications and no training on your data
We process your data under the GDPR, and your prompts and answers are never used to train any model. We do not store the content of your text, transcription or voice requests. GPU Solutions, the company behind GPU Flow, is certified ISO/IEC 27001:2022 and ENS Medium (Spain's national security framework). You get a Spanish invoice in euros; if your company is in the EU and has a valid VAT number, it is issued without VAT under the intra-EU reverse charge.
frequently asked
What developers ask before they start.
And if yours isn't here, write to us: a person answers, not a bot.
Is Token Factory compatible with the OpenAI API?
Yes. It exposes the same endpoints (/v1/chat/completions) and the Anthropic one (/v1/messages). Point the SDK you already use at https://api.gpuflow.ai/v1 (the Anthropic one at https://api.gpuflow.ai), paste your key, and the rest of your code stays the same.
Which open-source models can I call through the API?
Everything in the catalogue with one key: Qwen, GLM, Gemma, Nemotron, ALIA, Whisper, Qwen3-TTS and ACE-Step. If you need another open-source model, tell us.
How much does the LLM API cost?
From €0.06 per million input tokens, depending on the model. A million tokens is about 750,000 words. There are no monthly fees: you load credit and it is drawn down by actual use.
Is there a free tier?
Every new account gets €5 of Token Factory credit, enough for around 50,000 answers with Qwen3.5 9B. You activate it with a card.
Does it work with LangChain, LlamaIndex, Cursor or Claude Code?
Yes, with anything that supports a custom OpenAI or Anthropic base URL. Claude Code connects with environment variables; there is a step-by-step guide.
Do I need a developer to use it?
To connect it to your own website or app, yes, for an afternoon. If you have no developer, write to us and we will walk you through it.
Are there usage limits?
You consume against your prepaid balance: pay per token up to whatever you've topped up, no quotas or minimums. If you need sustained high volume, reach out and we'll raise your account limits.
Do you train on my data?
No. Your prompts and responses are processed to serve your request and are not used to train models. The details are in the privacy policy.
Do you offer an SLA?
We publish availability on the status page. For a contractual SLA (enterprise accounts), reach out and we'll set one up.
Start free with €5
Create your account, add your card, activate your credit and send your first request. No subscription and no minimum.