On November 12, AWS released the next-generation Amazon Bedrock AgentCore runtime, which its pricing page calls Runtime v2. Every agent session still runs in its own Firecracker microVM, but AWS says sessions now start up to 15x faster, in under 2 seconds regardless of image size or concurrency, and a single session can use up to 8 vCPUs, 32 GB of memory and a 10 GB container image. Memory is reclaimed while the agent waits on a model or tool, idle agents scale to zero, and multi-turn agents use up to 68% less memory per session. Sessions can also mount storage such as S3 as a filesystem so data stays inside the isolation boundary. v2 unit prices are higher than v1; the savings come from paying only for what is used.
Key Takeaways
- ✓Vendor claim: sessions start up to 15x faster, with cold starts under 2 seconds regardless of image size or concurrency, so warm pools are no longer needed
- ✓Per-session resources grow 4x: up to 8 vCPUs, 32 GB of memory and 10 GB container images; microVM sessions run for up to 8 hours
- ✓Idle agents scale to zero and memory is reclaimed during waits, cutting memory per session by up to 68% for multi-turn agents
- ✓v2 consumption prices: $0.1276 per vCPU-hour and $0.0169 per GB-hour, above v1 at $0.0895 and $0.00945; committed baseline is $0.0997 and $0.0132
- ✓Customer figures: Coinbase cut cold starts from nearly a minute to under 5 seconds; Zuora cut memory per invocation by about 60%
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and the problem
Agents now run for hours, executing code, driving browsers and touching sensitive data, so every session needs an isolated sandbox. The trade-off is hard: pre-provisioned environments cost money while idle, while on-demand ones add cold-start delay. Citing Deloitte, AWS says only 15% of organizations have scaled multi-agent systems and 70% cannot yet trust and govern their agents.
Architecture and how it works
The core guarantee stays: each session gets its own Firecracker microVM with its own identity, access policy and audit trail. Three things change. First, bigger and faster sessions: up to 8 vCPUs, 32 GB of memory and 10 GB images, starting in under 2 seconds. Second, usage-based billing: memory is reclaimed while the agent waits on a model, a tool or a human, and idle agents scale to zero; always-on agents can set a committed baseline and burst up to 4x above it. Third, isolation extends to data: sessions can mount an S3 bucket as a filesystem through FUSE, keeping data inside the boundary. AWS says per-session storage scoping will arrive by re:Invent.
Benchmarks and measured results
AWS published no standardized benchmark; all figures below are vendor claims: up to 15x faster starts, 4x more compute and up to 68% better memory efficiency. Customer reports: Coinbase cut cold starts from nearly a minute to under 5 seconds while weekly session volume more than doubled; Mission Cloud saw 35% faster inference cold starts and 17% lower cost per call; Zuora cut memory per invocation by about 60%. Note that the pricing page lists higher unit prices for v2 than v1, so savings depend on how much time an agent spends waiting.
Getting started
Teams already on AgentCore can run one multi-turn workload on both v1 and v2, compare cold starts and bills, then migrate. Large repo builds, browser automation and small local model inference can use the new 8 vCPU / 32 GB ceiling. Always-on support agents fit committed baseline pricing, and jobs longer than 8 hours should use Instances, which run for up to 14 days.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.