You can adopt these independently — Runtime without Memory, Gateway without Runtime.
Prices verified 2026-10-24 against the
AWS pricing page.
AgentCore harness
An agent from configuration
Declare the model, tools, skills and instructions; AgentCore supplies the environment, compute, memory, identity, networking and observability that turn that config into a running agent.
The harness itself is free, but it is not a way around the meters: everything it provisions bills at that component’s own rate. It also reaches fewer regions than Runtime — 15 of the 31 tracked here, with no N. California, Milan, Spain, Malaysia, Hyderabad or Thailand.
No separate charge; you pay only for the underlying resources it uses.
AgentCore Runtime
Serverless execution with per-session isolation
Runs your agent as a managed, serverless workload. Each session gets its own isolated microVM, and you are billed for CPU only while the agent is actually computing — not while it waits on a model response or an HTTP call.
Agents spend most of their wall-clock time waiting. Billing CPU on active use rather than session duration is the difference between paying for thinking and paying for waiting, and it is the single biggest reason Runtime is cheaper than holding a container open.
- per vCPU-hour (v2)
- $0.128
- per GB-hour (v2)
- $0.017
- per vCPU-hour (v1)
- $0.089
- per GB-hour (v1)
- $0.009
AWS lists two microVM rates and does not say which a new runtime gets. Its own pricing example uses v2, which also reclaims idle memory after 120 seconds, so the cost model here uses v2.
AgentCore Runtime Instances
Dedicated capacity instead of per-session microVMs
The reserved-capacity half of Runtime. Instead of a fresh microVM per session billed on active CPU, the agent runs on instances you hold, billed as the underlying EC2 cost plus an AgentCore management fee.
It inverts Runtime’s economics. microVMs bill only while the agent computes, which is why they are cheap for bursty work; instances bill while they exist. They win on steady, high-utilisation load and lose badly on anything idle — and they reach far fewer regions than microVMs do.
EC2 instance cost plus an AgentCore management fee: 12% standard, 7.8% on GPU instances.
AgentCore Gateway
Turns APIs and Lambdas into MCP tools
Exposes existing REST APIs, Lambda functions and third-party services to your agent as Model Context Protocol tools, with a semantic search index so the agent finds the right tool instead of being handed all of them.
Tool descriptions live in the prompt, so every unused tool is tokens you pay for on every single step. Semantic tool selection keeps that prefix small, which compounds hard in a loop.
- per 1,000 API invocations
- $0.005
- per 1,000 search queries
- $0.025
- per 100 tools indexed / month
- $0.020
- per GB of data egress to customer-owned VPCs
- $0.006
AgentCore Memory
Short-term and long-term recall
Managed conversation state. Short-term memory holds the working session; long-term memory extracts durable facts and retrieves them into later sessions.
Priced by event size and per record, not per token. Since 2026-10-06 short-term events bill per GB with a 12 KB floor and a 64 KB cap, so a long conversation costs more to remember than a short one, but the cap keeps any single event from becoming expensive.
- per GB short-term ingested
- $1.00
- per GB short-term retrieved
- $0.200
- per GB-month short-term stored
- $0.100
- per 1,000 long-term records stored / month
- $0.750
- per 1,000 long-term retrievals
- $0.500
- per 1,000 long-term records stored / month, override or self-managed strategy
- $0.250
Each short-term event bills at a 12 KB minimum and 64 KB maximum for ingestion and retrieval; storage bills actual size over the TTL. These meters replaced $0.25 per 1,000 events on 2026-10-06.
AgentCore Identity
OAuth and credentials on the agent's behalf
Brokers tokens so an agent can act against Google, Slack, GitHub or your own OAuth provider as a specific user, without the credentials passing through your prompt or your code.
Free when the agent reaches tools through Runtime or Gateway. Route authentication that way and this line disappears from the bill entirely.
- per 1,000 token or API-key requests
- $0.010
No charge when used through AgentCore Runtime or Gateway.
Built-in Tools
Browser and Code Interpreter
A managed headless browser for agents that operate real web UIs, and a sandboxed code interpreter for agents that need to compute rather than estimate.
Both bill per vCPU-hour and GB-hour while they run, so a browser session left open keeps costing until it is closed. Close them. The rate matches Runtime v1, not v2.
- per vCPU-hour
- $0.089
- per GB-hour
- $0.009
Web Search
Managed search for agents
A first-party search tool an agent can call without you contracting a third-party search API.
At $7 per 1,000 queries this is the most expensive per-call item in AgentCore by a wide margin. An agent that searches on every step will make search, not tokens, the largest line on the bill.
- per 1,000 queries
- $7.00
Policy in AgentCore
Rules on what the agent may do
Natural-language and structured authorization rules evaluated before an agent takes an action, so 'never issue a refund over €500 without approval' is enforced outside the model rather than asked of it.
The token component is billed separately from your model spend, and at $0.13 per 1,000 input tokens it is well above most model rates — keep policy prompts short.
- per authorization request
- $0.000025
- per 1,000 input tokens
- $0.130
The first 100 temporal policies per policy engine incur no authorization charge.
AgentCore Evaluations
Scoring agent trajectories
Judges whether an agent actually did the right thing across a whole multi-step trajectory, rather than whether one response looked plausible.
The failure mode of agents is a sequence that goes wrong in step four, which single-response evaluation cannot see.
- per 1,000 built-in input tokens
- $0.002
- per 1,000 built-in output tokens
- $0.012
- per 1,000 custom evaluations
- $1.50
- per 1,000 batch evaluation input tokens
- $0.002
- per 1,000 batch evaluation output tokens
- $0.009
AgentCore Observability
Traces of every step
OpenTelemetry-compatible traces of the agent's reasoning, tool calls and token usage, surfaced in CloudWatch.
Billed at CloudWatch rates rather than a separate AgentCore meter — the cost lands on a different line of your bill than the rest of this table.
Charged at standard CloudWatch rates; there is no separate AgentCore price.
AgentCore Optimization
Insights and A/B tests over agent runs
Analyses recorded agent runs and surfaces insights and recommendations, with A/B tests for comparing changed configurations against live traffic.
Insights and recommendations carry no AgentCore charge today, so the only cost of looking is the A/B tests — and those bill the model tokens and component usage they generate at standard rates, like any other traffic.
Insights (preview) and recommendations are free. A/B tests bill the model and component usage they generate at standard rates.
AWS Agent Registry
A catalogue of agents and their records
A registry of agent records that agents and operators can search, so an agent can discover another agent rather than being wired to it.
Both meters have a free allowance large enough that most teams never pay: 5,000 records and a million search calls a month. Reaching only 5 of the regions tracked here is the real constraint, not the price.
- per 1,000 records / month, after 5,000 free
- $0.400
- per 1,000 search calls, after 1M free
- $0.020
- per 1,000 List / Get API calls, after 2M free
- $0.004
AgentCore Payments
Lets an agent transact through a wallet provider
Wires an agent to a wallet provider so it can execute a payment as part of a task rather than handing the step back to a human.
There is no AgentCore meter on it at all — the cost is whatever your wallet provider charges. It is the one component on this page whose price does not appear on the AWS pricing page, so budget it from Coinbase or Stripe, not from here.
No AgentCore charge. Billed by the wallet provider (Coinbase or Stripe) under their own fees.