AWS Bedrock experiment cost $8.43 in one day with no results
Running Claude Code via Amazon Bedrock promised simpler billing and unlimited sessions, but one developer’s experiment ended with an $8.43 bill and no usable output in a single day. The lesson: AWS Bedrock charges for every API call, regardless of success, making it easy to run up costs during debugging or retries.
Behind the two ways to run Claude Code
Claude Code works with Anthropic’s direct API or through AWS Bedrock. The first route ties billing to an Anthropic subscription with session limits, while the second routes requests through AWS and charges via the customer’s account. From a user’s perspective, both feel identical, but the pricing model diverges sharply. Bedrock bills by tokens processed in each request, including failed attempts.
The experiment that went wrong
After configuring Claude Code to use Bedrock, the developer sent a detailed prompt expecting a quick result. Instead, repeated API errors appeared—timeouts and “unable to process” messages—along with retries that never produced usable output. Each attempt still incurred token charges, and with no success, the cumulative cost mounted quickly.
What the bill revealed
Two AWS budget alerts—at $5 and $10—triggered after just one day of experimentation. The Cost and Usage dashboard showed $8.43 for Claude Sonnet models on Bedrock, plus tax, for zero functional output. The monthly forecast jumped to $11.59, more than doubling the initial $5 budget. The key takeaway: Bedrock charges for compute time, not successful results, so debugging can become unexpectedly expensive.
Source: DEV Community. AI-assisted editorial synthesis — TechnoExpress.