ClickSorcery

API costs

What is a token — and what does it cost?

AI models measure work in tokens, not words or pages. Everything you need to know — without the jargon.

Token = a piece of text

A token is roughly ¾ of an English word — sometimes a whole word, sometimes half. Estonian uses slightly more tokens because compound words and case endings are longer.

Models read and write tokens — that's why their work is measured in tokens, not characters.

100,000 tokens is roughly

English text

150–300

A4 pages

depends on line spacing

Estonian text

120–200

A4 pages

Estonian uses ~20% more

Lines of code

2–3k

lines of code / JSON

≈ 70–100 pages

Rule of thumb1 million tokens ≈ 10 average books. A typical short chat uses 500–2,000 tokens — about one A4 page.

What does it mean in money?

AI providers bill per 1 million tokens. Here are the numbers in plain language:

Price per 1M tokens — most common models

ModelInput (you read)Output (it writes)What you get
Claude 3.5 SonnetBest quality$3.001M input tokens$15.001M output tokens~500 longer conversations
Gemini 2.0 FlashFast & cheap$0.101M input tokens$0.401M output tokens~5,000 conversations
DeepSeek V3Code & analysis$0.271M input tokens$1.101M output tokens~2,000 conversations
Llama 3.1 8B (Groq)Simple tasks$0.051M input tokens$0.081M output tokensGroq free tier: free
Example: One Sonnet conversation (500 input + 300 output tokens) costs about $0.006 — six cents. You get ~16 conversations for one euro.

How ClickSorcery keeps your bill low

ClickSorcery uses smart routing — each task goes to the best and cheapest capable model. The CEO handles critical decisions; simple agents use cheap models.

RoleModelPrice / 1MWhy
CEO / JerryCritical decisionsOpenRouter Claude 3.5 / 3.7 Sonnet$3 → $15in → outUsed less often — top quality only where it matters
C-Suite & analystsLogs, files, projectsGoogle / OpenRouter Gemini 1.5 / 2.0 Flash$0.10 → $0.40in → outHuge context, ultra-fast — a fraction of Sonnet pricing
Code specialistsDev, analysisDeepInfra DeepSeek V3$0.14 → $0.28in → outTop-tier code at a fraction of Claude cost
Light specialistsClassification, tutoringGroq Llama 3.1 8B$0.05 → $0.08in → outFree tier: 30 requests/min free

How a typical day splits

Most work runs on cheap models. Expensive models only show up at the most important moments.

80%

Simple work — free / nearly free

~€0

15%

Code / analysis

~€1–2

5%

CEO

~€1–2

Context cachingWhen an agent's system instructions are long, providers cache them on their side. Repeat reads cost 50–80% less — big savings on long-running chats.

€2–5

typical monthly cost with normal use (33 agents, reactive mode)

Agents don't run background work 24/7 — they respond when you need them.

ClickSorcery does not add a markup on API costs — you pay your provider directly.

← Back to landing