Amazon Bedrock AgentCore Runtime vs SF Compute Autoresearch Sandboxes

Amazon Bedrock AgentCore Runtime costs $0.0895 per vCPU-hour. SF Compute Autoresearch Sandboxes bills memory only, at $0.04572 per GiB-hour. Amazon Bedrock AgentCore Runtime runs microVMs and limits a session to 8 hours. SF Compute Autoresearch Sandboxes runs microVMs.

Updated Oct 23, 2026

Amazon Bedrock AgentCore Runtime

Hosts AI agents on AWS and runs each user session in its own microVM.

Isolation: MicroVM
Free tier: Not listed
Max session: 8 hours

Visit site Amazon Bedrock AgentCore Runtime profile

SF Compute Autoresearch Sandboxes

Isolated microVMs for AI agents that fork from a saved snapshot, part of SF Compute Autoresearch.

Isolation: MicroVM
Free tier: None
Max session: Not listed

Visit site SF Compute Autoresearch Sandboxes profile

What is different

7 fields differ. The rest are the same or not listed.

Field

Amazon Bedrock AgentCore Runtime

SF Compute Autoresearch Sandboxes

vCPU, per hour: Amazon Bedrock AgentCore Runtime: $0.0895; SF Compute Autoresearch Sandboxes: Included
RAM, per GiB-hour: Amazon Bedrock AgentCore Runtime: $0.00945; SF Compute Autoresearch Sandboxes: $0.04572
Storage: Amazon Bedrock AgentCore Runtime: Session storage is free during preview. Code storage is billed at Amazon ECR or S3 rates; SF Compute Autoresearch Sandboxes: Sandbox snapshots $0.10 per GiB-month of deduplicated bytes, $0.05 per GiB-month in cold storage
Billing: Amazon Bedrock AgentCore Runtime: Active CPU time only; SF Compute Autoresearch Sandboxes: Memory only
Largest sandbox: Amazon Bedrock AgentCore Runtime: 2 vCPU · 8 GiB; SF Compute Autoresearch Sandboxes: 4 vCPU · 8 GiB
Concurrent sandboxes: Amazon Bedrock AgentCore Runtime: Up to 5,000; SF Compute Autoresearch Sandboxes: Up to 4,096
Integrations: Amazon Bedrock AgentCore Runtime: Claude Agent SDK, LangChain, MCP, OpenAI Agents SDK; SF Compute Autoresearch Sandboxes: MCP, Claude Code

Full comparison

18 fields, at each provider's list price

Amazon Bedrock AgentCore Runtimeaws.amazon.com/bedrock/agentcore/pricingaws.amazon.com

SF Compute Autoresearch Sandboxesautoresearch.sfcompute.com

Pricing

vCPU, per hour: Amazon Bedrock AgentCore Runtime: $0.0895; SF Compute Autoresearch Sandboxes: Included
RAM, per GiB-hour: Amazon Bedrock AgentCore Runtime: $0.00945; SF Compute Autoresearch Sandboxes: $0.04572
Storage: Amazon Bedrock AgentCore Runtime: Session storage is free during preview. Code storage is billed at Amazon ECR or S3 rates; SF Compute Autoresearch Sandboxes: Sandbox snapshots $0.10 per GiB-month of deduplicated bytes, $0.05 per GiB-month in cold storage
Free tier: Amazon Bedrock AgentCore Runtime: Not listed; SF Compute Autoresearch Sandboxes: None
Paid plan from: Amazon Bedrock AgentCore Runtime: Pay as you go; SF Compute Autoresearch Sandboxes: Pay as you go
Billing: Amazon Bedrock AgentCore Runtime: Active CPU time only; SF Compute Autoresearch Sandboxes: Memory only

Limits

Max session: Amazon Bedrock AgentCore Runtime: 8 hours; SF Compute Autoresearch Sandboxes: Not listed
Idle stop: Amazon Bedrock AgentCore Runtime: After 15 minutes; SF Compute Autoresearch Sandboxes: Not listed
Largest sandbox: Amazon Bedrock AgentCore Runtime: 2 vCPU · 8 GiB; SF Compute Autoresearch Sandboxes: 4 vCPU · 8 GiB
Concurrent sandboxes: Amazon Bedrock AgentCore Runtime: Up to 5,000; SF Compute Autoresearch Sandboxes: Up to 4,096

Platform

Isolation: Amazon Bedrock AgentCore Runtime: MicroVM; SF Compute Autoresearch Sandboxes: MicroVM
GPU: Amazon Bedrock AgentCore Runtime: Not listed; SF Compute Autoresearch Sandboxes: Not listed
Snapshots: Amazon Bedrock AgentCore Runtime: Not listed; SF Compute Autoresearch Sandboxes:
Pause and resume: Amazon Bedrock AgentCore Runtime:; SF Compute Autoresearch Sandboxes:

Developer

SDKs: Amazon Bedrock AgentCore Runtime: Python, TypeScript; SF Compute Autoresearch Sandboxes: Not listed
Integrations: Amazon Bedrock AgentCore Runtime: Claude Agent SDK, LangChain, MCP, OpenAI Agents SDK; SF Compute Autoresearch Sandboxes: MCP, Claude Code

Trust

License: Amazon Bedrock AgentCore Runtime: Not listed; SF Compute Autoresearch Sandboxes: Not listed
BYOC: Amazon Bedrock AgentCore Runtime: Not listed; SF Compute Autoresearch Sandboxes: Not listed

Hourly cost

One sandbox running for one hour

Size

Amazon Bedrock AgentCore Runtime

SF Compute Autoresearch Sandboxes

1 vCPU · 2 GiB: Amazon Bedrock AgentCore Runtime: $0.108; SF Compute Autoresearch Sandboxes: $0.091
2 vCPU · 4 GiB: Amazon Bedrock AgentCore Runtime: $0.217; SF Compute Autoresearch Sandboxes: $0.183
4 vCPU · 8 GiB: Amazon Bedrock AgentCore Runtime: $0.434; SF Compute Autoresearch Sandboxes: $0.366

Compute only, at list rates.

Open the calculator

Which one fits

Based on the fields above

Amazon Bedrock AgentCore Runtime fits when

You use the Claude Agent SDK, LangChain or the OpenAI Agents SDK.
Your code waits a lot: CPU is billed only while it is in use.
You need SOC 2, HIPAA and ISO 27001.

SF Compute Autoresearch Sandboxes fits when

You use Claude Code.
You want the lower list price: $0.183 against $0.217 an hour for 2 vCPU and 4 GiB.
You need snapshots.
You want one rate for memory, with CPU included.

More comparisons

See all comparisons

Amazon Bedrock AgentCore Runtime

SF Compute Autoresearch Sandboxes

What is different

Full comparison

Pricing

Limits

Platform

Developer

Trust

Hourly cost

Which one fits

Amazon Bedrock AgentCore Runtime fits when

SF Compute Autoresearch Sandboxes fits when

More comparisons

Related

Compare

Pricing

Browse