Vendor cost breakdown
Amazon Bedrock cost in 2026 — and the capacity nobody uses
Amazon Bedrock is usage-based, not seat-based: it charges per input and output token, priced per model, across on-demand, batch, Flex, Priority, and provisioned modes. Because the spend lands inside the AWS bill next to everything else, it is the AI cost teams most often can’t attribute — and the one that quietly keeps billing from reserved capacity no one is using.
Prices verified against official vendor pages on Jun 12, 2026. All waste figures are estimates from your own inputs.
Example reconciliation
A typical setup
Billing
What you pay
$500/mo
Estimated from a typical setup — DevSpend reconciles these three the same way across every vendor. Edit it below.
Amazon Bedrock plans and per-seat prices
VENDOR| Plan | Price / seat / month | Best for | Usage notes |
|---|---|---|---|
| Bedrock on-demand | Usage-based | Usage-based (per token) | Pay per input/output token, priced per model. Standard, Flex (discounted, higher latency), and Priority (premium, faster) tiers. |
| Bedrock batch | Usage-based | Asynchronous bulk jobs | 50% off on-demand pricing for non-urgent workloads. |
| Bedrock provisioned / reserved | Usage-based | Guaranteed-throughput workloads | Fixed price for reserved tokens-per-minute capacity (1- or 3-month terms). |
Where teams waste money on Amazon Bedrock
TELEMETRYFINANCE- Bedrock has no permanent free tier — every API call bills from the first token, and the spend hides inside the broader AWS invoice unless it is tagged and broken out.
- Provisioned or reserved throughput keeps billing whether or not traffic arrives; capacity bought for a launch and forgotten is classic AWS-AI waste.
- Batch and Flex tiers cut token costs by up to 50% for non-urgent workloads — workloads left on Standard out of inertia pay double.
- Without cost-allocation tags or separate accounts, Bedrock spend can’t be attributed to teams — making it impossible to know which product or team owns the cost.
Estimate your Amazon Bedrock spend
The calculator below is prefilled with a typical Amazon Bedrock setup — for example, this one runs $500/month. Swap in your own seats and prices for a live estimate. Save it to a free workspace and DevSpend tracks the gap over time across every vendor; Pro adds per-user and per-team cost metrics and reconciliation against your real billing. Nothing is saved until you create a workspace.
Tool inputs
Seats, price, active estimate, and usage/API spend. Nothing is saved yet.
Active is your best estimate of seats in real use — the gap to assigned seats is your recoverable waste. Active can’t exceed seats, so it’s capped automatically.
Amazon Bedrock cost FAQ
How is Amazon Bedrock priced?
Bedrock is usage-based: you pay per input and output token at rates set per model, with no permanent free tier. On-demand is the default; batch processing costs 50% less for asynchronous jobs; Flex trades latency for a discount; Priority pays a premium for faster responses; and provisioned/reserved capacity bills a fixed price for guaranteed throughput on 1- or 3-month terms.
Why is AWS-billed AI spend hard to track?
Because it is buried in the AWS bill: Bedrock token charges appear as line items among hundreds of others, often across multiple accounts. Unless teams tag resources or use Cost Explorer filters, total AI spend on AWS stays invisible to the people deciding on seats and tools.
How should a team budget for Bedrock?
Treat it like cloud infrastructure, not a subscription: estimate monthly token volume per workload, choose batch or Flex tiers for anything non-urgent, set AWS budget alerts on the Bedrock service, and review provisioned capacity monthly. Track the resulting monthly number alongside seat-based AI subscriptions — a DevSpend AI workspace keeps both in one view.
How is Bedrock different from Kiro on the AWS bill?
Both bill through AWS, but differently: Bedrock charges per token for model inference, while Kiro is a per-developer credit subscription for AWS’s agentic IDE. They leak in different ways — Bedrock from forgotten provisioned capacity and wrong tiers, Kiro from over-provisioned plans and idle developers — so DevSpend tracks them as separate vendors.
Track every vendor, not just the spend you can see
Or run every AI tool through one free workspace — spend by vendor, paid seats vs active users, and your waste number, always visible (up to 5 subscriptions free). Pro adds unlimited tracking, per-user and per-team cost metrics, expected-vs-billed reconciliation, and a print-ready reclaim report. See Pro — $39/mo.