Cost model

Up to 24× cheaper per evidenced answer.

The same model, the same question: alone with manually pasted documents – or with c:node, which passes on only the evidenced citations. All prices, assumptions and sources disclosed.

ModelAlone (light) · with c:node (dark)AloneWith c:nodeFactor
Cost per evidenced answer in euros, list prices as of 29 Sep 2026. “Alone” = best case with prompt caching. Time spent gathering documents is not included.
Assumptions

How it is calculated.

Every figure can be recalculated.

Model alone:node + model
Calls until the answerQuestion and three follow-ups – each round resends the entire history.
4
1
Context per call · contract caseAlone: 23 contracts plus statutory excerpts pasted in manually. With c:node: only the evidenced citations.
175,000 tokens
6,000 tokens
Output tokens · large modelsVisible answer plus reasoning. With c:node the context is already evidenced – less reasoning effort.
900 + 1,500
800 + 400
Tokens for evidence checkingThe quote must appear verbatim in the source – c:node checks this deterministically, without a model.
no check
0deterministic
Best case for “alone”: active prompt caching – context written once (+25%), then read cheaply€1 = US$1.1378 · ECB reference rate 28 Sep 2026Net · all amounts excluding VAT
Sources

Where the figures come from.

Every price links to the provider's page.

Anthropic – Claude model pricing

Opus 5.5, Sonnet 5.5, Haiku 4.5 · list prices per 1M tokens, incl. cache prices

anthropic.com
OpenAI – API pricing

GPT-6 Sol and GPT-6 Luna · list prices per 1M tokens

developers.openai.com
Google – Gemini API pricing

Gemini 3.1 Pro and 3.8 Flash · up to 200,000 tokens of context

ai.google.dev
Mistral – API pricing

Mistral Large 3 and Small 4 · calculated without cache discount

mistral.ai
5
AWS – EC2 g4dn.xlarge (NVIDIA T4)

US$0.658 per hour, on-demand, eu-central-1 · our own invoice 09/2026

aws.amazon.com
Qwen 2.5 7B – model card

Estimated throughput on T4: 1,500 tokens/s input, 35 tokens/s output, 4 parallel requests

huggingface.co
7
European Central Bank – reference rates

€1 = US$1.1378 on 28 Sep 2026

ecb.europa.eu
8
Context volume per case

Our own estimate from the example cases, e.g. 23 contracts of around 7,000 tokens each plus excerpts from the Data Act and GDPR

Our own assumption

List prices retrieved on 29 Sep 2026. All amounts net, converted to euros.

Calculate with your numbers.

We set up the model with your case numbers and volumes.

Book a call View pricing
FAQ

About the cost model.

Why is c:node cheaper if it uses the same model?

Because the model doesn't have to read entire folders. c:node finds the relevant passages beforehand and passes on only those. That saves input tokens – and follow-up rounds.

Are the figures measured or calculated?

Calculated – with list prices and disclosed assumptions. The throughput of open models on your own GPU is estimated, not measured.

What about the time spent gathering documents?

That is not included. Without c:node, someone has to find and paste in the documents – with c:node, the connectors do that. So in practice the savings are larger.

Why is the factor so high for Mistral?

In our calculation, Mistral grants no cache discount. Alone, the full context is paid in full every round – with c:node, only the evidence, once.