How Much Does AWS Bedrock Cost?
Bedrock bills per token by model — and the bill depends entirely on which models your agents call and how many times. Here's the structure and how to control it.
How Bedrock pricing works
Per-token pricing varies by model family and size — frontier models cost multiples of compact ones.
Provisioned throughput (reserved capacity) is priced by hour; on-demand by token.
Your bill = model choice × request volume. Agents multiply both.
The hidden cost: agent request patterns
Retry loops re-send the same request — same tokens, charged again.
Multi-agent fan-out multiplies inference volume across parallel branches.
Big context windows (long documents, tool outputs) are expensive per call.
What it really costs
Model mix + request volume + context size. Teams that control all three control the bill.
The incident database shows how unmonitored agents turn small per-token costs into large ones.
Where the money goes
| Lever | What it controls | Tool |
|---|---|---|
| Model selection | Per-token price | Route to cheaper models |
| Request volume | Token count | Velocity limits, caps |
| Context size | Tokens per call | Context trimming, caps |
| Merchant scope | Which vendors | Merchant allowlist |
FAQ
Is Bedrock cheaper than direct APIs?
It depends on the model and whether you use on-demand or provisioned throughput. Compare current per-token rates for your model mix.
What drives Bedrock cost most?
Frontier models at volume. A single agent calling a frontier model in a loop is the fastest way to a large bill.
Can sipi.bot control Bedrock spend?
Yes — category rules and per-agent caps govern inference spend, and velocity limits kill retry loops before they multiply.
Related
Stop the next $12,400 night.
One API call (or MCP tool) in front of every agent transaction — APPROVED, BLOCKED, or FLAGGED, deterministic, ~5 ms, fully logged.
See plans — from $99/mo Try a live check