Claude Cowork, priced honestly.
We couldn’t find a reliable resource for calculating and comparing token costs across real business workloads. So we built this calculator for our clients.
Per-token rates on the three platforms are effectively identical. Differences come from EU data residency, cloud support, and the capabilities and constraints that come with each.
Pick a preset, change the settings and the future trajectory with a click. Share this page’s link - it carries exactly your model.
Claude Cowork for your 500-person mid-sized DACH enterprise needing EU-only data residency
After 12 months of adoption.
Available on AWS Bedrock and GCP Vertex AI,
see all 3 compared ↓
Your 12-month adoption journey
Refine assumptions
Repeated context is cached at up to 90% off. Higher hit rate = lower cost.
Knowledge workers average 18 to 22 active days.
Opus is 1.7x Sonnet. Big lever for Power users.
USD-set pricing, EUR shown at 1:1.
Compare the three platforms
Providers that do not meet your requirements are struck through.
| AWS Bedrock | GCP Vertex AI | Anthropic direct | |
|---|---|---|---|
| Cost · monthly | |||
| Inference | — | — | — |
| Cloud support (Business+ / Enhanced) | — | — | — |
| Anthropic seat fee | — | — | — |
| Total estimate | — | — | — |
| Per user / month | — | — | — |
| Compliance | |||
| EU data residency | Geo CRIS (eu.*) | eu multi-region | — |
| BSI C5 (Germany) | — | ||
| ISO 27001 / 27017 / 27018 | all three | all three | 27001 only |
| HIPAA BAA | via AWS | via Google | Enterprise only |
| CLOUD Act exposure | Yes | Yes | Yes |
| What you get | |||
| Cowork desktop app | |||
| Routines (scheduled agents) | — | — | |
| Skills Marketplace | — | — | |
| Computer Use (operate other apps) | |||
| Web search tool | — | ||
| Code execution tool | — | — | |
| Batch API (50% off async workloads) | — | ||
| Reliability | |||
| Published SLA | 99.9% | 99.5% | 99.99% (Enterprise) |
| Actual 90-day uptime | ~99.9% | ~99.9% | 98.6-99.4% |
| Provisioned Throughput | GA, published rates | Private Preview | Sales-only |
What you give up for EU data residency:
These features ship only on Anthropic direct (1p Cowork), which has no EU data residency option in 2026. Switch the residency requirement to "Anywhere" to keep them.
Vendor placement
Where each option fits.
AWS Bedrock
The default if your organisation already runs on AWS. Same per-token rates as Vertex, same EU residency story, AWS’s full compliance stack (BSI C5, HIPAA, ISO 27001) flows through. If you grow beyond ~€20K/month, you can lock in predictable pricing through AWS’s reserved-capacity programme.
If you are not already on AWS, the GCP option is functionally equivalent.
GCP Vertex AI
The default if your organisation already runs on Google Cloud. Same per-token rates as Bedrock, same compliance posture in practice, and one cleaner EU endpoint. The real win is if your data already lives in BigQuery: Claude can read it without leaving Google’s tenancy.
If you are not already on GCP, the AWS option is functionally equivalent.
Anthropic direct
The shortest path to Claude: no cloud-provider procurement, pay Anthropic directly. Newest features land here first (Routines, Skills, Computer Use). $20 per seat per month on top of usage.
The hard disqualifier for most DACH enterprises: no EU data residency option in 2026. Workable for US-headquartered teams, pilots, or workloads with no personal data.
Methodology. Per-token pricing from AWS Bedrock, GCP Vertex AI, and Anthropic (May 2026). Tier mix from Anthropic’s Enterprise Consumption Guide. 10% EU surcharge per Anthropic’s cloud-platform policy. Cloud support modeled symmetrically at the AWS Business+ / GCP Enhanced tier (about $100 floor plus a tiered percentage of monthly inference spend). EDP or CUD customers will land lower.
Caching. Modeled. Anthropic’s prompt caching cuts repeated-context input costs by up to 90%. The calculator assumes a typical enterprise hit rate per preset (40 to 75%). Adjust under Refine assumptions if your workload is more or less cache-friendly.
±20% covers real-world variance from cache hit rate, model mix, output ratio, and concurrent-user patterns.
Need help reading the output, negotiating a cloud contract, or designing the actual rollout? Book a call.