Blog · Published · Updated
Claude API Pricing 2026: Token Costs and a Real Budget
Short answer
The Claude API costs from $0.10 per million input tokens and $0.50 output on Haiku 5.5 to $10 and $50 on Fable 5.1. Sonnet 5.5 is $2 and $10, Opus 5.5 $4 and $20. At 10,000 tasks of 2,000 input tokens and 500 output tokens, the monthly bill runs from $4.50 on Haiku to $450 on Fable, and on Sonnet it is $90, or €80.
How we workDetails ↓
- 01Call30 minutes with an engineer, free
- 02PrototypeA working system on your process and data, free
- 03ImplementationTarget and fixed price in the contract; 50% at signing
- 04Acceptance50% once the agreed target is met
- analysis with an engineer and a free prototype
- €0
- team in ChatGPT or Claude, up to 10 people
- from €1,150
- automation, agent or application for one process
- from €3,000
We belong to the OpenAI and Anthropic partner programmes and learn through their official courses. What this means ↓
Author: Syntalith · 3 min read
Statistics
The model API market: statistics 2026
$90
a month for 10,000 tasks on Claude Sonnet 5.5, at 2,000 input tokens and 500 output tokens each, or €80. With the fixed instructions cached, about $61.50. Scenario.
40%
According to Menlo Ventures (2025), Anthropic holds 40% of large companies' spending on language model APIs and OpenAI 27%, based on a survey of about 500 US decision makers.
Source: Menlo Ventures, State of Generative AI in the Enterprise (2025) ↗
40%
According to Gartner (2025), over 40% of AI agent projects will be cancelled by the end of 2027 because of costs, unclear value or weak risk control.
50%
According to the European Commission (2026), half of EU SMEs and mid-caps name high investment costs as a barrier to AI.
Source: European Commission, Flash Eurobarometer 578 (2026) ↗
14–56%
According to Anthropic (2025), studies with a control group found that AI cut task time by 14% to 56%.
Source: Anthropic, Estimating AI productivity gains (2025) ↗
Figures checked: 22 October 2026
The current Claude API price list
| Model | Input per 1 million | Output per 1 million | What for |
|---|---|---|---|
| Claude Fable 5.1 | $10 | $50 | Hard reasoning, long analyses, agentic work over many steps |
| Claude Opus 5.5 | $4 | $20 | Complex cases and agents working across several systems |
| Claude Sonnet 5.5 | $2 | $10 | Production routine: replies, extraction, document flow |
| Claude Haiku 5.5 | $0.10 | $0.50 | Classification, routing cases, short high-volume tasks |
Rates from Anthropic's pricing, checked on 22 October 2026. Haiku 5.5 arrived on 7 October 2026, and on the same day Anthropic cut the cache read rate on Sonnet 5.5 to $0.10 (Haiku 5.5 announcement). The input rate differs a hundredfold between Haiku 5.5 and Fable 5.1, and output costs five times more than input on every model.
What 10,000 tasks cost
2,000 input tokens and 500 output tokens per task, no cache and no retries. Converted at the ECB rate of 1.1205 USD per euro on 22 October 2026.
Źródło: Anthropic, API pricing (2026) ↗
- Base volume
- 10,000 × (2,000 × $2 + 500 × $10) / 1,000,000 = $90
- Retrying every tenth task
- $90 × 1.10 = $99
- Token cost per task
- $0.0099
$99 a month, or €88
The task sizes and the share of retries are the scenario's assumptions. The time of the person who checks a sample of results and the exceptions, upkeep of the integration and taxes are separate items. Converted at the ECB rate of 1.1205 USD per euro on 22 October 2026.
Do not guess the tokens per case. Run a few dozen real cases through the model and read the usage from the API response. The system instructions, the history and the documents go into the input on every call, and an agent that takes several steps pays for input at every step.
How to lower the Claude API bill
A model per step
Haiku for routing
Haiku 5.5 classifies and routes, Sonnet handles the routine, Opus and Fable only the exceptions.
Cache
$0.10 per read
Fixed instructions cached on Sonnet 5.5 cost twenty times less than ordinary input.
Batch
half the rate
Tasks that do not need an answer within seconds: overnight, reports, enriching a database.
If 1,500 of the 2,000 input tokens are the same instructions read from cache, 10,000 tasks on Sonnet 5.5 cost about $61.50 in place of $90. Output stays the same, so shorter answers cut the bill more than a change of model. Set daily and monthly limits before the start and decide what happens when the budget runs out: a queue, a cheaper model or an alert to a person.
Tokens are not the whole cost
In a typical rollout the API bill is the smallest item. The rest is the process analysis, integrations, tests, launch and maintenance.
Process analysis
30 min, written plan within 2 working days
€0
An engineer counts the cost of the manual work and shows what to do first.
Test on your process
5 working days
€0
The system works on 20 to 100 cases your team has already handled.
Audit
5 working days
from €600
A process map, the architecture and a fixed price. If you order the build, we credit it towards the rollout.
ChatGPT or Claude for the team
2 weeks
€1,150–4,600
Accounts, rules for working with data and training. Up to 10, up to 40 and above 40 people.
AI automation
2–6 weeks
from €3,000
A process with fixed steps: inbox, documents, CRM, reports.
AI app
4–10 weeks
€3,000–7,000
A tool for the team, for example search across documents with the source.
AI agent
6–16 weeks
€3,000–17,500
Does the routine steps on its own, hands exceptions to a person, records everything.
Claude is also available through the clouds of the big providers, each with its own price list and terms, and the route you choose affects data residency. At a small volume of a few people working with documents, a subscription is cheaper and faster, as the article on Claude Team and Enterprise describes. Book a free process analysis, and the summary will give you an estimate of the model cost and the scope of the rollout.
