Claude API Pricing EU 2026: Bedrock, Google, Foundry
Claude API pricing in the EU: Fable 5.1, Opus 5, Sonnet 5 and Haiku 4.5 on Anthropic, Bedrock, Google Cloud and Foundry, the 10% EU premium and 3 cost examples.
TL;DR
Claude API pricing in the EU is the global list price plus 10%. Sonnet 5 costs 2 / 10 USD per million input / output tokens globally, 2.20 / 11 USD on Bedrock EU or Google Cloud EU. The Anthropic API and Microsoft Foundry offer no EU-only processing. A 200-person chatbot on Sonnet 5 costs 434 to 572 USD a month.
What does the Claude API cost per model and platform?
On 19 September 2026, global list prices were identical everywhere: Fable 5.1 at 10 / 50 USD, Opus 5 at 5 / 25, Sonnet 5 at 2 / 10 and Haiku 4.5 at 1 / 5 USD per million input / output tokens. Bedrock and Google EU endpoints add 10%. No vendor publishes EUR prices.
Figures come from the Anthropic pricing page, the Amazon Bedrock pricing page (EU Frankfurt and EU Ireland selectors, price file published 11 September 2026) and the Google Cloud Claude pricing table. These are pay-per-token rates; seat-based subscriptions from Pro to Team are covered in our Claude pricing plans overview.
| Platform and endpoint | Fable 5.1 | Opus 5 | Sonnet 5 | Haiku 4.5 | EU processing? |
|---|---|---|---|---|---|
| Anthropic API (global) | 10 / 50 | 5 / 25 | 2 / 10 | 1 / 5 | No |
| Bedrock global CRIS | 10 / 50 | 5 / 25 | 2 / 10 | 1 / 5 | No |
| Bedrock EU geo or in-region | not offered | 5.50 / 27.50 | 2.20 / 11 | 1.10 / 5.50 | Yes |
| Google Cloud global | 10 / 50 | 5 / 25 | 2 / 10 | 1 / 5 | No |
| Google Cloud EU multi-region | 11 / 55 | 5.50 / 27.50 | 2.20 / 11 | europe-west1 only: 1.10 / 5.50 | Yes |
| Foundry Global Standard | 10 / 50 | 5 / 25 | 2 / 10 | 1 / 5 | No |
USD per million tokens, input / output. Foundry bills in Claude Consumption Units at 0.01 USD each, with token rates equal to the Claude API. Sonnet 5’s 2 / 10 price is now standard: Anthropic states that the increase to 3 / 15 USD planned for 1 September 2026 will not happen.
Where does the 10% EU premium apply?
The 10% premium applies to every token category on Bedrock’s EU geo and in-region endpoints and on Google Cloud’s EU multi-region and regional endpoints. It is the only way to buy Claude with EU-bound processing today. The Anthropic API and Foundry are cheaper in theory, but neither offers an EU processing option you could pay for.
Anthropic’s first-party API supports only global and us as inference_geo, and the US option costs 1.1x. Foundry offers Global Standard and a US Data Zone, and Microsoft lists the EU Data Zone for Claude as not available. For a European company, the practical choice is therefore Bedrock EU or Google EU, both at the same premium. Our Amazon Bedrock guide shows which non-Claude models Bedrock also offers in the EU, and our Kimi K3 Bedrock analysis prices two that do not. Our view: 10% is a small line item next to a transfer impact assessment. Details per platform are in our GDPR comparison of the four Claude providers.
Two exceptions matter. Fable 5.1 has no EU price on Bedrock at all, so in Europe it runs only on Google’s EU multi-region at 11 / 55 USD. Haiku 4.5 is missing from Google’s EU multi-region tab and appears only under europe-west1.
How much do batch and prompt caching save?
Batch halves the price of input and output tokens, and cache reads cost 10% of the input price (2.5% on Fable 5.1). Both discounts stack with the EU premium. For workloads with long, repeated context, caching saves more than choosing a cheaper model, and it makes the EU premium irrelevant.
| Model | Anthropic API batch | Bedrock EU batch | Google EU batch | Cache read EU (Bedrock / Google) |
|---|---|---|---|---|
| Fable 5.1 | 5 / 25 | not offered | 5.50 / 27.50 | n/a / 0.275 |
| Opus 5 | 2.50 / 12.50 | 2.75 / 13.75 | 2.75 / 13.75 | 0.55 / 0.55 |
| Sonnet 5 | 1 / 5 | not listed (N/A) | 1.10 / 5.50 | 0.22 / 0.22 |
| Haiku 4.5 | 0.50 / 2.50 | 0.55 / 2.75 | 0.55 / 2.75 (europe-west1) | 0.11 / 0.11 |
Cache writes cost more than normal input: 1.25x for a 5-minute cache and 2x for a 1-hour cache. On the Sonnet 5 EU rate that is 2.75 and 4.40 USD. Foundry does not support the Message Batches API, so batch discounts are not available there. On Bedrock, Sonnet 5 shows no batch price, which pushes large Sonnet 5 batch jobs towards Google’s EU endpoint.
Why could your token budget be 30% too low?
Claude 4.7 and later models, which includes Fable 5.1, Opus 5 and Sonnet 5 but not Haiku 4.5, use a newer tokenizer that produces about 30% more tokens for the same text. A budget built from word counts or from an older Sonnet 4.x bill will undershoot. The exact increase depends on the content, so measure with your own data.
Anthropic states this on its pricing page. Run a representative sample through the token counting endpoint before you commit a budget. Sonnet 5 at 2 USD per million input tokens is still cheaper than Sonnet 4.6 at 3 USD, even with 30% more tokens.
What do three typical EU workloads cost per month?
Three realistic workloads land between about 80 and 7,200 USD per month on EU endpoints. The assumptions below are ours, not vendor data. Token counts already assume the new tokenizer. EUR values use the ECB reference rate of 18 September 2026, 1 EUR = 1.1460 USD, and ignore VAT and negotiated discounts.
| Example | Assumptions | Model and endpoint | USD / month | EUR / month |
|---|---|---|---|---|
| Internal chatbot, 200 employees | 10 requests per person and working day, 20 days: 40,000 requests. 4,000 input tokens (2,000 system prompt) and 500 output tokens each | Sonnet 5, EU, no caching | 572 | 499 |
| Same chatbot | System prompt cached, 90% cache hits, 10% 5-minute cache writes | Sonnet 5, EU, with caching | 434 | 379 |
| Document pipeline, 50,000 pages | 1,500 input and 300 output tokens per page, overnight run | Sonnet 5, Google EU batch | 165 | 144 |
| Same pipeline | Same volume, no batch price on Bedrock | Sonnet 5, Bedrock EU on-demand | 330 | 288 |
| Same pipeline | Same volume | Haiku 4.5, Bedrock EU batch | 82.50 | 72 |
| Coding assistant, 20 developers | 150 requests per developer and day, 20 days: 60,000 requests. 40,000 input tokens (90% cache reads, 10% cache writes), 1,000 output tokens | Sonnet 5, EU | 1,795 | 1,566 |
| Same assistant | Same | Opus 5, EU | 4,488 | 3,916 |
| Same assistant | Same | Fable 5.1, Google EU | 7,194 | 6,277 |
Three conclusions. First, the chatbot is cheap: under 3 USD per employee per month. Second, the pipeline price depends more on the platform than on the EU premium, because Bedrock lists no Sonnet 5 batch price. Haiku 4.5 halves the cost again, but its Anthropic retirement is not before 15 October 2026, so we would not build a new pipeline on it. Third, coding agents are the real budget item. Without caching, the Sonnet 5 variant would cost 5,940 USD instead of 1,795 USD. Setup details are in our guides for Claude on AWS Bedrock in the EU and Claude on Vertex AI in Europe.
FAQ
How much does the Claude API cost per month for a mid-sized EU company?
For a typical internal chatbot with 200 users on Sonnet 5 via an EU endpoint, expect roughly 430 to 570 USD per month, depending on prompt caching. A coding assistant for 20 developers costs about 1,800 USD on Sonnet 5 and 4,500 USD on Opus 5. Those are our assumption-based estimates, not vendor quotes. If staff need the Claude apps with seats instead, our Claude Enterprise vs Bedrock comparison sets both cost models side by side.
Bedrock EU vs Google Cloud EU: which is cheaper for Claude?
On-demand prices are identical: Sonnet 5 costs 2.20 / 11 USD and Opus 5 5.50 / 27.50 USD on both. Differences appear elsewhere. Google lists a Sonnet 5 batch price in the EU multi-region (1.10 / 5.50 USD) and offers Fable 5.1 in the EU, while Bedrock lists neither.
Is there an official EUR price for Claude?
No. Anthropic states that all payments are in USD, and AWS and Google publish Claude prices in USD. Your cloud invoice may be issued in EUR, but the conversion follows your provider’s billing terms. Our EUR figures use the ECB reference rate of 1.1460 USD per EUR from 18 September 2026.
Does the Anthropic API cost less than Bedrock for EU companies?
The list price equals Bedrock global, and Sonnet 5 batch costs 1 / 5 USD there. But the Anthropic API has no EU processing option, only global and us. If your data must be processed in the EU, the relevant comparison is Bedrock EU or Google EU at a 10% premium. Our Claude API key guide explains when a direct Anthropic key still makes sense.
How much does prompt caching save on Claude?
A cache read costs 10% of the input price, or 2.5% on Fable 5.1. In our coding assistant example, caching cuts the Sonnet 5 EU bill from 5,940 to about 1,795 USD per month. Caching pays off after one read for a 5-minute cache and after two reads for a 1-hour cache.
Sources
- Anthropic docs: Pricing (19 September 2026)
- Amazon Bedrock pricing (19 September 2026)
- Google Cloud: Agent Platform pricing (Claude models) (19 September 2026)
- Anthropic docs: Claude in Microsoft Foundry (19 September 2026)
- Microsoft Learn: Foundry models from partners (18 September 2026)
- Anthropic docs: Data residency (18 September 2026)
- ECB euro foreign exchange reference rates (19 September 2026)