Data as of 16/11/2026

AI model pricing at SEION

AI at SEION

What it costs to use artificial intelligence at SEION, how each usage is calculated, and the rules behind pricing.

How usage is measured

Tokens

Language models charge by tokens, small pieces of text (one token is roughly four characters, or about three-quarters of a word). Each model call has:

Input

What is sent to the model: agent instructions, conversation, documents, and tool results.

Cached input

The part of the input the model received recently (fixed agent instructions or conversation history). It is usually charged at 10% of the input price.

Output

What the model generates: its answer and internal reasoning. It is the most expensive component and usually weighs most in the bill.

Input categories are mutually exclusive: each token is charged only once, at its applicable rate.

Agents, assistants, automations

Text models

Prices per million tokens.

GPT-6 family

ModelBest forInputCachedOutputTier
gpt-6-astraComplex reasoning, programming, demanding professional work20.002.00100.00€€€€
gpt-6.1-solHigh capability, balancing intelligence and cost4.000.4020.00€€€
gpt-6-lunaEfficient for focused tasks, automation, and high volume0.200.021.00€

GPT-5.6 family

ModelBest forInputCachedOutputTier
gpt-5.6-solComplex professional work, agents, advanced reasoning8.000.8040.00€€€
gpt-5.6-terraBalanced intelligence, speed, and cost4.000.4024.00€€€
gpt-5.6-lunaLow cost, high speed, high volume0.400.042.40€€

GPT-5 family

ModelBest forInputCachedOutputTier
gpt-5.5Larger context and agent capabilities10.001.0060.00€€€€
gpt-5.4Maximum context and performance for advanced agents5.000.5030.00€€€
gpt-5.3Advanced reasoning and complex tasks3.500.3528.00€€€
gpt-5.2Complex analysis and deep reasoning3.500.3528.00€€€
gpt-5.1Business automation and assistants2.500.2520.00€€€
gpt-5-miniFast, low-cost chat and repetitive tasks0.500.054.00€€
gpt-5-nanoSimple tasks and high volume0.100.010.80€

GPT-4o family

ModelBest forInputCachedOutputTier
gpt-4oVersatile multimodal: support, creativity, visual analysis5.002.5020.00€€€
gpt-4o-miniAffordable lightweight automation and basic support0.300.151.20€

GPT-4o cached input is 50% of input pricing; for other models, 10%.

Images

Image generation

ModelInput (text and source images)Cached inputOutput (generated image)
gpt-image-28.002.0030.00

Note: Prices per million tokens. Input uses the image rate because source images make up most of the input when editing or composing.

Approximate cost per image by quality and size

Quality1024×10241024×1536 / 1536×1024
Low0.0080.012
Medium0.0320.047
High0.1250.187

Approximate amounts per image; actual cost depends on the tokens generated.

Models without individual pricing

Default pricing

If usage comes from a model without its own price, the generic rate applies (per million tokens):

Input5.00
Cached input0.50
Output20.00

Note: A model is charged at its own rate or the generic rate, never another model’s rate.

For every AI call

How each usage charge is calculated

01

Select the price.

Use the model’s price; if unavailable, the agent’s configured model price; otherwise, the default.

02

Count the tokens.

Separate input, cached input, and output.

03

Apply the minimum.

Every text call counts at least 8 input tokens.

04

Cost = tokens × rate.

Calculate each component; the total is their sum.

05

No cache rate.

If the model has no cached-input rate, cached tokens are not charged.

Each usage record stores its tokens, itemized cost, and applied rate. Later price changes do not affect historical usage.

Example

One gpt-5.6-terra turn

10,000 input tokens (6,000 cached) and 800 response tokens.

With gpt-6-luna it would cost 0.0017; with gpt-6-astra, 0.1720.

ItemTokensRate / 1MCost
Input4,0004.000.0160
Cached input6,0000.400.0024
Output80024.000.0192
Total0.0376

Model catalog

Price tier (€ to €€€€)

In SEION’s model catalog, each model displays a tier from 1 to 4 for quick comparison:

  • It is calculated from the sum of input and output rates per token, preserving the order of output pricing, which dominates the bill.
  • The cheapest model is tier 1 and the most expensive is tier 4; the rest are distributed logarithmically because prices span two orders of magnitude.
  • It is recalculated from current prices and changes when rates change.
  • Transcription and tools (priced per minute or unit) are not assigned a tier.
€Tier 1
€€Tier 2
€€€Tier 3
€€€€Tier 4

Spending control

Per-agent usage limits

Set a maximum spend for each agent, combining selected usage periods (today, this month, by contact, etc.) and optionally applying a condition. Once exceeded, the agent stops responding and actions can be triggered, such as notifying someone.

Visible usage

Each agent shows its spend and tokens for today, the current month, and the previous month. The AI usage log details every call.