Agents cost more than tokens
Bedrock AgentCore Pricing Calculator
Estimate microVM or Runtime Instances, Memory, Gateway, and Web Search — budget the platform before model tokens.
- No signup
- Instant results
- ~4 min
Rates reviewed
AgentCore components
us-east-1list rates, pricing as of 2026-10-17. Model inference billed separately. microVM charges active CPU and peak memory only. Runtime Instances add EC2 On-Demand plus a management fee. Source: AWS public pricing.
Runtime
microVM bills active CPU and peak memory. Runtime Instances bill the EC2 On-Demand price for the hours the instance runs, plus a management fee on that On-Demand rate.
Active CPU time only — LLM I/O wait is not billed.
Charged from boot until stop. The management fee is 12% of this On-Demand total, or 7.8% for G-series.
Used by microVM Runtime, Browser, and Code Interpreter. Runtime Instances ignore this.
128 MB minimum on microVM. Runtime Instances ignore this.
Gateway, Memory, Web Search, Policy
Short-term events and long-term records are different quantities. Checking a component on step 1 applies these amounts; leaving it off ignores them.
Gateway
$0.005 per 1,000 invocations.
$0.025 per 1,000 searches.
$0.02 per 100 tools/month.
Memory
Since Oct 6, 2026: $1.00/GB ingested + $0.20/GB retrieved (12 KB min, 64 KB max per event) + $0.10/GB-month stored. Assumes each event is read once and kept 30 days.
AWS's example uses 3 KB events, billed at the 12 KB minimum for ingestion and retrieval.
Stored records, billed hourly on a 31-day month. AWS example uses 10k records against 100k short-term events.
$0.50 per 1,000 retrievals.
Web Search and Policy
$7 per 1,000 queries.
$0.000025 per request. The first 100 temporal policies per engine have no extra authorization charge — not modeled here.
$0.13 per 1,000 tokens when natural language is converted to Cedar.
Browser, Code Interpreter, Evaluations, Registry
Browser and Code Interpreter
Evaluations
$0.0024 per 1,000 input tokens. Batch evaluations are 25% lower and are not included.
$0.012 per 1,000 output tokens.
$1.50 per 1,000 custom evaluations. Model usage for custom evaluators is extra.
Agent Registry
First 5,000 records each month are free, then $0.40 per 1,000.
First 1,000,000 calls free, then $0.020 per 1,000.
First 2,000,000 combined calls free, then $0.004 per 1,000.
Estimated AgentCore Platform Cost
Platform fees excluding model inference
Total Monthly Cost
live resultPlatform vs inference split unlocks after email
Unlock platform vs inference budget guidance
See memory retention tips and link to Bedrock token pricing for full agent TCO.
Full breakdown unlocked
TCO guidance is now visible.
Want this applied to your real account?
FactualMinds designs production agents on Amazon Bedrock AgentCore — approval boundaries, IAM tool scopes, and cost-aware model routing.
Talk to GenAI Experts →Who This Tool Is For
AI platform teams, CTOs, and architects budgeting production agents with autonomous tool access — beyond token inference alone.
Why We Built This Tool
AgentCore bills Runtime, Gateway, Memory, Web Search, and Policy separately from Bedrock model tokens. A tool-heavy agent can accumulate platform fees before inference dominates.
What Problem It Solves
- Platform vs inference split. Budget AgentCore Runtime and Gateway alongside token costs.
- Two memory meters. Short-term events and long-term stored records are different quantities.
- Runtime Instances. EC2 hours plus a 12% management fee (7.8% on G-series) replace the microVM meter for long-lived sessions.
Pair with our Bedrock Pricing Calculator, GenAI Readiness Assessment, or Generative AI on AWS service.
Frequently Asked Questions
How is microVM Runtime priced?
microVM Runtime charges $0.0895 per vCPU-hour and $0.00945 per GB-hour on active consumption, with a 128 MB memory minimum. I/O wait during LLM or tool calls is not billed as CPU time. These are the V1 active-consumption rates, not Runtime V2 elastic memory.
How are Runtime Instances priced?
Runtime Instances bill the underlying EC2 On-Demand price for the hours the instance runs, plus an AgentCore management fee of 12% of that On-Demand amount (7.8% for G-series). Savings Plans and Reserved Instances apply to the EC2 portion in your account, not to the management fee. This calculator uses list On-Demand only.
Is AgentCore separate from Bedrock model inference?
Yes. AgentCore covers the agent platform. Bedrock foundation model token costs are billed separately — use the Bedrock pricing calculator for inference. Identity has no extra charge when used through Runtime or Gateway.
How does Memory billing work?
Since October 6, 2026, short-term memory bills $1.00 per GB ingested, $0.20 per GB retrieved (each event counted at 12–64 KB), and $0.10 per GB-month stored — 100,000 3 KB events cost about $1.47 a month. Before that date it was $0.25 per 1,000 events. Long-term storage is $0.75 per 1,000 stored records per month on built-in strategies, or $0.25 on override and self-managed strategies. Long-term records never expire by default; since October 13, 2026 you can set a TTL (minimum 31 days) per strategy so the stored-record count stops growing. Retrievals are $0.50 per 1,000. Do not put the same number in both event and record fields — AWS bills them as different quantities.
What do Web Search, Policy, Evaluations, and Registry cost?
Web Search is $7 per 1,000 queries. Policy is $0.000025 per authorization request and $0.13 per 1,000 authoring tokens. Built-in evaluations are $0.0024 per 1,000 input tokens and $0.012 per 1,000 output tokens; custom evaluations are $1.50 per 1,000. Registry is free for the first 5,000 records, 1,000,000 Search calls, and 2,000,000 List/Get calls each month.
How accurate is this estimate?
Rates are AWS us-east-1 list prices as of 2026-10-17. Payments (Coinbase or Stripe wallet operations) and Runtime V2 elastic memory are not included. Verify current rates on the AgentCore pricing page before production budgeting.