Cross-platform decision guide · kept current
Where to run your Claude agent: API vs Bedrock vs Cloudflare
It's not a price decision — Claude's per-token cost is identical on the Anthropic API and AWS Bedrock. It's a decision about where you want the abstraction boundary to sit: do you want Anthropic to manage the agent runtime, do you want it inside your own AWS account next to your IAM and CloudWatch, or do you want it at the edge close to users?
Rough rule: Anthropic API / Managed Agents for the fastest path and newest features; Bedrock when you're already on AWS and want consolidated IAM/billing/governance; Cloudflare for low-latency, globally distributed, edge-native workloads. Details and current pricing below.
The decision in one table
| Option | Choose it when | Cost shape | Watch out for |
|---|---|---|---|
| Anthropic API / Claude Managed Agents | You want the simplest path, the newest models first, and Anthropic to own the agent harness/session lifecycle | Standard Claude token rates; Managed Agents add ~$0.08 per active session-hour | Less native integration with your existing cloud IAM/observability |
| AWS Bedrock / AgentCore | You already run on AWS (IAM, VPC, CloudWatch, consolidated billing) and want governance in one place | Identical Claude token rates; AgentCore Runtime bills for active consumption per second, I/O wait free (~$0.0895/vCPU-hour + ~$0.00945/GB-hour); other AgentCore services priced separately | New Claude models/features tend to land on the direct API first, with Bedrock catching up within weeks |
| Cloudflare Workers AI | You need global edge inference, low latency for user-facing apps, and data-residency by default | Its own model catalogue and metering (per-model rates, roughly $0.10–$1.40 / 1M input tokens depending on model) | A different model catalogue than first-party Claude; use the AI Gateway to route to Claude where needed |
Prices change. These reflect public pricing as of June 2026 and must be re-checked against each provider's pricing page before you rely on them.
The thing nobody states plainly: token price is not the variable
Because Claude's per-token pricing is identical on the Anthropic API and Bedrock, the platform choice rarely turns on the model bill. It turns on operational fit: where your identity, secrets, networking, logging, and billing already live, and how much of the agent runtime you want to operate yourself versus hand to the provider. That framing — the abstraction boundary — is the spine of the decision, and it's exactly the kind of call a senior cloud/infra engineer is already equipped to make.
A note on lock-in
The cleanest hedge is to keep the agent logic and tool layer portable and put a routing layer (such as an AI Gateway) between your code and the provider, so switching where inference runs is a config change, not a rewrite. That's also good cost-observability hygiene. Don't over-engineer it before you have a reason to switch — but design the boundary so you can.
Frequently asked
Is Claude cheaper on Bedrock or on the Anthropic API?
Per-token pricing for Claude is the same on both as of 2026. Choose on operational fit (where your IAM/observability/billing live and how new you need the model versions), not on token price.
What do Claude Managed Agents cost on top of tokens?
Roughly $0.08 per active session-hour in addition to normal token rates — you pay for the time the managed agent runtime is running. Verify on Anthropic's current pricing page.
Can I run the same Claude model on Cloudflare?
Cloudflare Workers AI has its own model catalogue. To use Claude specifically at the edge, route through the Cloudflare AI Gateway to Anthropic — you get edge proximity plus first-party Claude, with caching and cost observability in the gateway.
Which should I learn first?
Learn the portable layer first — the agentic loop, tools, and evals — because it transfers across all three. Then learn the deployment specifics of whichever platform your target employers use. The integrated path teaches all three deliberately.
- Anthropic — Claude API pricing; Managed Agents add ~$0.08/active session-hour (public beta since Apr 2026).
- AWS — Amazon Bedrock AgentCore pricing (Runtime $0.0895/vCPU-hour + $0.00945/GB-hour, active-consumption billing) and Bedrock model pricing (Claude token rates matching the direct API).
- Cloudflare — Workers AI pricing (its own per-model catalogue; current text models run roughly $0.10–$1.40 / 1M input tokens). Abstraction-boundary framing also draws on cross-platform comparisons (DEV.to AWS Builders, CloudZero, Anchor Sprint, 2026).
The Anthropic, AgentCore Runtime, Managed Agents, and Workers AI figures were verified against the providers' own pricing pages on 16 Jun 2026. Pricing is volatile — re-check the provider pages before relying on these. Corrections: hello@aiarch.dev.
Learn to deploy on all three — and justify the choice.
AI Architect Academy teaches the Anthropic, AWS, and Cloudflare deployment surfaces as one cross-platform architect competency — including the cost-modeling and routing that actually move the bill.
Get notified when new tracks ship.