There is no Claude Code licence to buy on Amazon Bedrock. You pay AWS for the tokens the agent consumes, at Bedrock's published per-token rates for whichever Claude model you point it at, and it lands on your normal AWS invoice. Anthropic publishes what that works out to in practice: "the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users." A Claude Pro seat is $17 a month. Read those two numbers next to each other before you decide Bedrock is the cheap route, because it is not the cheap route. It is the route where the money comes out of a budget you already have.
Every figure and quotation in this article was read from Anthropic's own Claude Code documentation and pricing page in August 2026. Nothing here is estimated.
What you are actually buying
Claude Code runs three ways, and they are billed in three different shapes. On a Claude subscription you buy a seat and the usage comes with it. Through the Claude Console you buy tokens directly from Anthropic. Through a cloud provider, which for most teams means Amazon Bedrock, you buy tokens from your cloud vendor and Anthropic never invoices you at all.
| Route | What you pay | Is there a ceiling? | Who invoices you |
|---|---|---|---|
| Claude Pro | $17 a month on the annual plan, $20 monthly | Yes, a usage window | Anthropic |
| Claude Max | From $100 a month | Yes, a larger usage window | Anthropic |
| Claude Team | $20 a seat annually ($25 monthly), or $100 a premium seat ($125 monthly) | Yes, per seat | Anthropic |
| Claude Enterprise | "$20 per seat per month plus usage billed at API rates" | No, the seat is a floor | Anthropic |
| Claude Console | Per token, at Anthropic's API rates | Workspace spend limits, if you set them | Anthropic |
| Amazon Bedrock | Per token, at Bedrock's rates for the model you pick | Only your own AWS budget controls | AWS |
One line in Anthropic's setup documentation rules out the cheapest option people ask about first: "Claude Code requires a Pro, Max, Team, Enterprise, or Console account. The free Claude.ai plan does not include Claude Code access." There is no free tier here at all. The entry point is $17 a month.
What Anthropic says a developer costs
This is the number to plan against, and it is unusual that a vendor publishes it. From Anthropic's cost management documentation: across enterprise deployments the average is around $13 per developer per active day and $150 to $250 per developer per month, and 90% of users stay below $30 per active day.
Three things follow from that, and the third is the one that catches teams out.
First, "active day" is doing real work in that sentence. A developer who uses the agent nine days in a month is not a $150 developer, they are a $117 developer. Your cost per head tracks adoption more than it tracks headcount, which means a pilot that goes well gets more expensive, not less.
Second, the range is wide because the inputs are wide. Anthropic attributes the variance to model selection, codebase size and usage patterns such as running multiple instances or automation. A team that pins Opus for everything and runs parallel sessions is at the top of that band or above it.
Third, and this is the part worth putting in front of whoever signs the invoice: 90% below $30 an active day means 10% are above it. That is not a criticism of the tool, it is the arithmetic of a metered product. On a $17 Pro seat the worst month is $17. On Bedrock there is no worst month, there is only a distribution. Anthropic's own advice is the right advice: "start with a small pilot group and use the tracking tools below to establish a baseline before wider rollout."
The thing the AWS bill will not tell you
Here is the practical problem nobody warns you about until you are three months in. Anthropic's documentation maps each deployment route to where you see spend, where you cap it, and how you pull per-user numbers. On a Claude for Teams or Enterprise plan you get a spend report CSV and, on Enterprise, an analytics API. On the Console you get a usage page and a Claude Code analytics API. On Amazon Bedrock you get this: spend appears in "your cloud billing console", you cap it with "your cloud's budget controls", and per-user reporting comes from OpenTelemetry or an LLM gateway.
In other words, Bedrock gives you a Bedrock line on your AWS bill. It does not tell you which developer produced it. If you want to know whether the spend is three people doing serious work or one person leaving an agent running, you have to build that yourself. Anthropic is direct about the only route that covers every setup: OpenTelemetry export "is the only option that streams per-user token and cost metrics into your own observability stack in near real time."
Budget for that work. It is a day or two of platform engineering, it needs to exist before rollout rather than after, and it is the difference between managing this cost and discovering it. Teams that already centralise cloud and SaaS spend in one place will find this easier, and if you do not have that yet, a read-only view of where your cloud and SaaS spend actually goes is the sort of thing that pays for itself the first time a metered vendor line moves without warning.
Two features you give up on Bedrock
Both are documented, both are small, and both surprise people who assumed the Bedrock deployment was simply the same product on different plumbing.
Web search does not work. Anthropic states it plainly: "The WebSearch tool is not available on Amazon Bedrock." If your developers have got used to the agent looking up a library's current API before it writes against it, that habit stops at the Bedrock boundary.
You cannot log out. "When using Amazon Bedrock, the /logout command is unavailable since authentication is handled through AWS credentials." Minor day to day, occasionally annoying on a shared machine, and worth knowing before someone files it as a bug.
Pin your model versions before you roll this out
This one is a genuine trap and Anthropic flags it with a warning box. Deploy Claude Code on Bedrock to a team without pinning model versions and the aliases resolve to Claude Code's built-in Bedrock defaults, "which can lag the newest release and may not yet be available in your account." Concretely: without setting ANTHROPIC_DEFAULT_SONNET_MODEL, the sonnet alias on Bedrock resolves to Sonnet 4.5. Without ANTHROPIC_DEFAULT_OPUS_MODEL, opus resolves to Opus 5.
So a developer on a Claude subscription and a developer on your Bedrock deployment can type the same word, get different models, and reasonably conclude that the tool got worse after the migration. Pin the versions and you control when your team moves. Anthropic also notes that Claude Code falls back to an earlier or lower-tier model at startup when the default is unavailable in your account, which is a quieter version of the same failure.
While you are in the setup: model access needs approval before any of this works, and if you use AWS Organizations you can submit the request once from the management account through the PutUseCaseForModelAccess API, with approval extending to child accounts automatically. That saves a lot of ticket-chasing on a large estate.
So when is Bedrock the right call?
Three situations, and price is not one of them.
Your data cannot leave your AWS account. This is the real reason most regulated teams run Claude Code on Bedrock, and it is a good reason. The traffic stays inside your account, under your existing agreements, in a region you chose. If you have already done the work of drawing that perimeter for your databases and your logs, keeping the coding agent inside it is consistent rather than paranoid.
You have committed AWS spend to burn down. An enterprise discount programme turns a new vendor line into an existing one, and procurement stops being a six-week conversation. This is a genuinely good reason and it is the one people are usually too embarrassed to say out loud.
You need unattended or CI use. Bedrock supports an API key through AWS_BEARER_TOKEN_BEDROCK, alongside the standard credential chain, SSO profiles and cross-account credential export. If the agent has to run without a human at a browser, this is the documented path.
If none of those three apply, buy seats. A Team seat at $20 a month with a spend report and a built-in ceiling is a better deal than $150 to $250 of metered tokens with neither, and it is a lot less platform work.
How this compares to buying an agent on a flat plan
It is worth putting the alternatives next to each other, because "Claude Code on Bedrock" competes less with other Anthropic plans than it does with the fixed-price agents.
| Option | Monthly cost per developer | Ceiling | Who can see per-user spend |
|---|---|---|---|
| Claude Code on Bedrock | $150 to $250 (Anthropic's published range) | None by default | Nobody, until you wire up OpenTelemetry |
| Claude Team seat | $20, or $100 for a premium seat | Yes | Admins, via the spend report CSV |
| Kiro Pro to Power | $20 to $200 | Yes, overage is off by default | Each developer, in the subscription dashboard |
| Agentcode | $29, flat | Yes, it is the price | Not applicable, there is nothing to meter |
If your AWS shop is weighing the Bedrock route against AWS's own agent, that is a real comparison and we have worked it through in detail on Kiro vs Claude Code, including the part that decides it for most teams: Kiro's number is a ceiling and Anthropic's is an average. For the plan grid on its own, Claude Code pricing covers every tier, and Kiro pricing does the same on the other side. If Amazon Q Developer seats are what you are migrating off, Amazon Q Developer against Kiro covers that path.
Frequently asked questions
How much does Claude Code cost on Bedrock?
There is no licence fee. You pay AWS for tokens at Bedrock's published rates for the Claude model you use. Anthropic's own documentation puts the practical result at around $13 per developer per active day and $150 to $250 per developer per month across enterprise deployments, with 90% of users staying under $30 per active day.
Is Claude Code on Bedrock cheaper than a subscription?
Usually no, and it is not sold as a saving. A Claude Pro seat is $17 a month and a Team seat is $20, against Anthropic's published $150 to $250 a month for metered deployments. Bedrock is chosen for data residency, existing AWS commitments and unattended use, not for price.
Do I need a Claude subscription to use Claude Code on Bedrock?
No. Bedrock is an alternative to a subscription, not an addition to one. You authenticate with AWS credentials, an SSO profile or a Bedrock API key, and Anthropic never bills you. Note that the /logout command is unavailable in this mode, because authentication comes from AWS.
Can I cap what Claude Code spends on Bedrock?
Only with your own AWS budget controls. Anthropic's documentation is explicit that on a cloud provider you cap spend with "your cloud's budget controls" rather than anything inside Claude Code. Set the budget and the alerts before you roll out, not after the first invoice.
Which Claude model does Bedrock use by default?
Whatever Claude Code's built-in Bedrock default resolves to, which can lag the newest release. Unpinned, sonnet resolves to Sonnet 4.5 and opus resolves to Opus 5. Anthropic recommends pinning specific model IDs before any multi-user deployment so you decide when your team moves.
Does everything in Claude Code work on Bedrock?
Almost. Two documented exceptions: the WebSearch tool is not available on Amazon Bedrock at all, and the /logout command is unavailable because AWS handles authentication. Everything else behaves as it does on a subscription.
The shape of the decision
Bedrock does not make Claude Code cheaper. It makes it yours: your account, your region, your agreement, your budget controls. That is worth real money to a regulated team and worth nothing at all to a five-person startup, which is why the same deployment gets recommended and dismissed by people who are both right.
What it definitely does is move the cost from a line you can predict to a line you have to watch. If you go this route, wire up the per-user telemetry in the same sprint you wire up the credentials. If watching a variable line item is not something your team has the appetite for, the honest answer is that a fixed price exists for a reason, and paying it is not a failure of ambition.
That is the trade we built Agentcode around. You connect a GitHub or GitLab repository, describe the task, and it plans, edits, runs your tests and opens a pull request a human reviews. It never merges on its own. It is $29 a month, flat, with nothing to meter and no month that surprises you. Narrower than Claude Code, deliberately, and priced like a subscription rather than a utility.
Keep reading