Billing · 4 min read
AI capacity and pricing
How token consumption is priced, the per-model rate table, and where to control cost.
When a model runs, it consumes tokens. How many depends on how much material went into the request and how long the response is — this can’t be predicted exactly in advance. Roughly 95% of total token consumption is input tokens, with the remaining 5% output.
How tokens are counted
Section titled “How tokens are counted”Different models price input and output tokens differently, so the same task can cost different amounts depending on which model handles it.
Pricing per million tokens
Section titled “Pricing per million tokens”Prices are exclusive of VAT and stated in Euro.
| Model type | Creator | Model | EUR / input M | EUR / output M |
|---|---|---|---|---|
| Closed source | OpenAI | GPT-5.5 | 4.6 | 27.7 |
| Closed source | OpenAI | GPT-5.4 | 2.3 | 13.8 |
| Closed source | OpenAI | GPT-5.2 | 1.6 | 12.9 |
| Closed source | OpenAI | GPT-5 | 1.2 | 9.2 |
| Closed source | OpenAI | GPT-5 mini | 0.2 | 1.8 |
| Closed source | OpenAI | GPT-4.1 | 1.8 | 7.4 |
| Closed source | OpenAI | GPT-4.1 mini | 0.4 | 1.5 |
| Closed source | OpenAI | GPT-4.1 nano | 0.1 | 0.4 |
| Closed source | OpenAI | GPT-4o | 2.3 | 9.2 |
| Closed source | OpenAI | GPT-4o mini | 0.1 | 0.6 |
| Closed source | OpenAI | GPT-5.6 Sol (Azure) | 4.7 | 28.3 |
| Closed source | OpenAI | GPT-5.6 Terra (Azure) | 1.9 | 11.3 |
| Closed source | OpenAI | GPT-5.6 Luna (Azure) | 0.2 | 1.1 |
| Closed source | OpenAI | GPT-5.5 (Azure) | 4.8 | 29.0 |
| Closed source | OpenAI | GPT-5.1 (Azure) | 1.2 | 9.2 |
| Closed source | OpenAI | GPT-5 mini (Azure) | 0.2 | 1.9 |
| Closed source | OpenAI | GPT-5 nano (Azure) | 0.1 | 0.4 |
| Closed source | OpenAI | GPT-4.1 mini (Azure) | 0.4 | 1.5 |
| Closed source | OpenAI | GPT-4.1 nano (Azure) | 0.1 | 0.4 |
| Closed source | Anthropic | Claude 4.8 Opus | 4.6 | 23.1 |
| Closed source | Anthropic | Claude 4.7 Opus | 4.6 | 23.1 |
| Closed source | Anthropic | Claude 4.6 Opus | 4.6 | 23.1 |
| Closed source | Anthropic | Claude 4.6 Sonnet | 2.8 | 13.8 |
| Closed source | Anthropic | Claude 4.5 Sonnet | 2.8 | 13.8 |
| Closed source | Anthropic | Claude 4.5 Haiku | 0.9 | 4.6 |
| Closed source | Anthropic | Claude 5.5 Opus (EU) | 4.1 | 20.3 |
| Closed source | Anthropic | Claude 5 Opus (EU) | 5.1 | 25.4 |
| Closed source | Anthropic | Claude 4.6 Opus (EU) | 5.1 | 25.4 |
| Closed source | Anthropic | Claude 5.5 Sonnet (EU) | 2.0 | 10.2 |
| Closed source | Anthropic | Claude 5 Sonnet (EU) | 2.0 | 10.2 |
| Closed source | Anthropic | Claude 4.6 Sonnet (EU) | 3.0 | 15.2 |
| Closed source | Anthropic | Claude 5.5 Haiku (EU) | 0.1 | 0.5 |
| Closed source | Anthropic | Claude 4.5 Haiku (EU) | 1.0 | 5.1 |
| Closed source | Gemini 4 Argon | 1.8 | 9.2 | |
| Closed source | Gemini 3.8 Flash | 0.7 | 3.5 | |
| Closed source | Gemini 3.7 Flash | 0.7 | 3.5 | |
| Closed source | Gemini 3.6 Flash | 0.7 | 3.5 | |
| Closed source | Gemini 3.5 Flash | 1.4 | 8.3 | |
| Closed source | Gemini 3.5 Flash-Lite | 0.3 | 2.3 | |
| Closed source | Gemini 3.1 Pro Preview | 1.8 | 11.1 | |
| Closed source | Gemini 2.5 Pro | 1.2 | 9.2 | |
| Closed source | Gemini 2.5 Flash | 0.3 | 2.3 | |
| Open source | Gemma 4 | 0.3 | 0.5 | |
| Open source | Mistral | Mistral 4 Large | 0.6 | 1.9 |
| Open source | Mistral | Mistral 4 Small | 0.1 | 0.6 |
| Open source | Mistral | Mistral 3.5 Medium | 1.4 | 6.9 |
| Open source | Mistral | Mistral 3 Large | 0.5 | 1.4 |
| Open source | Alibaba | Qwen3.8 27B | 0.6 | 3.3 |
| Open source | Alibaba | Qwen3.6 35B | 0.3 | 1.5 |
| Open source | Alibaba | Qwen3.5 397B | 0.6 | 3.6 |
| Open source | Moonshot AI | Kimi K3 | 3.0 | 15.0 |
| Open source | Swedish Hosted - Managed by Intric | 0.0 | 0.0 | |
| Closed source | OpenAI | Whisper | 0.3 | — |
| Open source | National Library of Sweden | KB - Whisper * | 0.1 | — |
For transcription, the price per query corresponds to 60 minutes of recording.
Long requests
Section titled “Long requests”Some models charge a higher rate once the input of a single request passes a threshold. When it does, Intric bills the whole request — input and output — at the higher rate. Every request in a conversation carries the earlier messages with it, so a long conversation or a large attached document can cross the threshold.
| Model | Threshold, input tokens per request | EUR / input M | EUR / output M |
|---|---|---|---|
| GPT-5.5 | 272,000 | 9.2 | 41.5 |
| GPT-5.4 | 272,000 | 4.6 | 20.8 |
| GPT-5.6 Sol (Azure) | 272,000 | 9.4 | 42.5 |
| GPT-5.6 Terra (Azure) | 272,000 | 3.8 | 17.0 |
| GPT-5.6 Luna (Azure) | 272,000 | 0.4 | 1.7 |
| GPT-5.5 (Azure) | 272,000 | 9.4 | 42.5 |
| Gemini 3.1 Pro Preview | 200,000 | 3.7 | 16.6 |
| Gemini 2.5 Pro | 200,000 | 2.3 | 13.8 |
| Claude 5.5 Haiku (EU) | 100,000 | 0.5 | 2.5 |
Image generation
Section titled “Image generation”Image models are priced per million tokens too: your prompt counts as text input, an image you supply counts as image input, and the generated image counts as image output.
| Creator | Model | EUR / text input M | EUR / image input M | EUR / image output M |
|---|---|---|---|---|
| OpenAI | GPT Image 2.5 Sunburst | 4.6 | 7.4 | 27.7 |
| OpenAI | GPT Image 2.5 Flare | 4.6 | 7.4 | 27.7 |
| OpenAI | GPT Image 2 | 4.6 | 7.4 | 27.7 |
| Gemini 2.5 Flash Image | 0.3 | 0.3 | 27.7 |
Custom and self-hosted models
Section titled “Custom and self-hosted models”Intric doesn’t automatically know the price of a custom or self-hosted model, so an administrator sets its token cost directly on the model’s configuration. See Models for where that’s configured, and keep the value current — an outdated or missing cost leads to under- or over-reported figures for that model.
How model choice affects cost
Section titled “How model choice affects cost”Choosing a more capable, more expensive model for a simple task costs more without necessarily improving the result. Reserve the highest-cost models for tasks that genuinely need their reasoning ability, and use an economical or fast option for everyday work — see Completion model for how that choice is made per assistant.
Controlling cost
Section titled “Controlling cost”To keep spending in check, use Billing groups: group users by cost center, set a limit per user, and grant extra credit where it’s needed.