Sandbox pricing that pays for active compute, nothing on standby
Per-second and per-gigabyte rates are roughly the same across the sandbox category. What changes the bill is the architecture. Because Blaxel scales sandboxes to zero the second they are idle and resumes the full memory state in about 25 milliseconds, teams run up to 70 percent more cost-effectively in practice at the same unit rates. The cause is infrastructure designed to scale to zero and resume instantly.
Pay as-you-go
Usage-based pricing with infinite scalability.
$0.00USAGE
- Up to $200 in free credits
- Access to all Blaxel products
- Tier system for higher limits, up to 100,000+ sandboxes
- Full observability suite
- Community support on Discord
Start for free
Custom
For dedicated and custom deployments.
Custom pricing deployments. Contact us for a quote.
- All features of Pay-as-you-go, plus:
- Up to 256 GB RAM per sandbox or job
- Dedicated deployment locations
- Private network connectivity
- Connect servers from your own infra
Talk to us
Trusted by Polsia, Webflow, Shortwave, and hundreds of other autonomous agent systems.
$0 of compute costs on standby
Idle sandboxes cost no active CPU. The runtime auto-suspends when idle, and suspended time never counts toward compute. You pay only for marginal snapshot storage time.
Same unit rates, lower bills
Up to 70 percent more cost-effective in practice through scale-down and consumption optimization, not discounted list prices.
Start free
Up to $200 in free credits. Spin up your first sandbox in under a second.
Scale with quotas
Grow to 50,000+ concurrent sandboxes and up to 512 TB of volume storage on a tier-based quota system that upgrades automatically as you top up.
[RATES]
Rate table
Compute
Give a computer to your agent. Persistent environments that idle to zero, resume in about 25 milliseconds, and scale instantly to 50,000+ concurrent machines.
Compute
Give a computer to your agent. Persistent environments that idle to zero, resume in about 25 milliseconds, and scale instantly to 50,000+ concurrent machines.
Sandboxes
One microVM per agent, app, or job. Boot in milliseconds, resume from standby in about 25 milliseconds, and keep full state intact.
| Resource / Item | Description | Rate |
|---|---|---|
| Active CPU | Time the sandbox spends actively running, billed by allocated memory (CPU scales automatically). Suspended time does not count. | $0.0000115 / GB RAM per second |
| Snapshots storage | Memory and filesystem snapshots that keep suspended sandboxes ready to resume. | $0.20 / GB per month |
| Images | Templated deployment images stored for fast, repeatable creation. | $0.045 / GB per month |
Batch API
Spawn thousands of tasks across individual, isolated sandboxes in seconds on an elastic backbone.
| Resource / Item | Description | Rate |
|---|---|---|
| Compute runtime | Active compute across every task in a job, billed by the memory it uses. | $0.000006 / GB RAM per second |
| Cron jobs | Schedule recurring jobs on a cron expression. | Included |
| Images | Templated deployment images. | $0.045 / GB per month |
Agent runtime (coming soon)
A session-first, fully stateful runtime for long-running agents that run for hours and hand off cleanly.
| Resource / Item | Description | Rate |
|---|---|---|
| Agent compute runtime | Active compute while an agent session runs, billed by the memory it uses. | Coming soon |
| Images | Templated deployment images. | Coming soon |
MCP hosting
Run MCP servers as first-class workloads on the same private network as your agents. Tool calls never leave Blaxel.
| Resource / Item | Description | Rate |
|---|---|---|
| Compute runtime | Active compute while an MCP server runs, billed by the memory it uses. | $0.000007 / GB RAM per second |
| Images | Templated deployment images. | $0.045 / GB per month |
Storage
Give your agents lasting memory: Agent Drive for multi-agent collaboration, Volumes for durable mounts that retain data for years, and local snapshots kept until deleted.
Storage
Give your agents lasting memory: Agent Drive for multi-agent collaboration, Volumes for durable mounts that retain data for years, and local snapshots kept until deleted.
Storage
Attach state that survives restarts, resumes, redeploys, and multi-agent collaboration.
| Resource / Item | Description | Rate |
|---|---|---|
| Standby snapshot storage | Memory and filesystem snapshots that keep a suspended sandbox ready to resume in about 25 milliseconds. | $0.20 / GB per month |
| Volumes | Durable block storage mounted into sandboxes. Retains data for years, fully redundant. | Usage-based |
| Images | Templated deployment images stored for fast, repeatable creation. | $0.045 / GB per month |
| Agent Drive | Distributed file system shared across sandboxes for multi-agent collaboration. | Private preview |
Network
Control everything agents can reach: outbound allow-lists, static IPs, proxy routing with secret injection, and a co-located model gateway.
Network
Control everything agents can reach: outbound allow-lists, static IPs, proxy routing with secret injection, and a co-located model gateway.
Networking
Shape egress, inject secrets, pin IPs, and route model calls from the same private agent backbone.
| Resource / Item | Description | Rate |
|---|---|---|
| Outbound control | Allow-list outbound traffic per workload, enforced below the runtime. | Included |
| Proxy routing | Managed egress proxy that injects secrets into HTTPS traffic and audits every request. | Included |
| Static IPs | Dedicated outbound IP addresses for your sandboxes. | Contact us |
| Model gateway | One co-located endpoint for every model provider. | Provider rates, see docs |
[QUOTAS]
Quotas and tiers
Scale up to 50,000+ concurrent sandboxes
Every account starts with free-tier quotas. As you top up credits, your tier automatically increases. View all tiers in the console.
| Tier | Minimum Top-up Required | Sandbox Limit |
|---|---|---|
| Tier 0 | Free | 10 |
| Tier 1 | $20 / month | 50 |
| Tier 2 | $50 / month | 200 |
| … 7 more tiers … | - | - |
| Tier 9 | Contact us | 100,000+ |
Tiers upgrade automatically with top-ups.
[SECTION]
Your journey
Traditional sandboxes provision from scratch on every run, stay alive for an hour at most, then lose their state forever. Blaxel inverts that. A Blaxel sandbox runs for any duration, drops to standby at $0 compute when idle, and restores in about 25 milliseconds on the next call - with process state, memory, and filesystem unchanged.
- 1
Start free
Begin with free credits to test. No credit card required.
- 2
Top up credits
When you hit limits, top up additional credits to increase your quota tier.
- 3
Pay what you use
Only pay for your usage, down to the second.
[SECTION]
Optional add-ons
Enhance your Pay-as-you-go plan with premium features and support.
Email support
Private email support with 24-hour response time during business hours. $800 per month, plus 3% of usage
Live Slack support
A direct Slack channel with Blaxel engineers for architecture advice and support, so you're never waiting for a response when you need it. $1,600 per month, plus 10% of usage
HIPAA Compliance
Get a signed Business Associate Agreement (BAA) to use Blaxel in a HIPAA-compliant manner. $250 per month
[SECTION]
Multiple deployment options
Run Blaxel the way your security and residency requirements demand.
The same runtime, the same 25 millisecond resumes, wherever your workloads need to live.
Managed cloud
The fastest path to production. Fully managed by Blaxel, with tier-based quotas that scale as you grow. Reach 50,000+ concurrent sandboxes.
Private networking
Run on Blaxel's managed cloud offering but interconnect your virtual private cloud through private endpoints, so your data never transits the public internet.
Bring-your-own-metal
The closest option for self-hosted deployments. Deploy bare metal servers in your own VPC, and privately connect them to Blaxel control plane for the sandbox runtime, so data stays on your hardware during execution.
Get a demo
Building at enterprise scale? Volume pricing, data residency, and self-hosting or VPC deployment are available. Talk to us about BYOC, dedicated deployment locations, and custom SLAs.
[FAQ]
Frequently Asked Questions
- How fast does an agent resume after being idle?
- Sandboxes scale to zero after about 5 seconds of inactivity and resume from standby in roughly 25ms, even after weeks of being suspended.
- Does my agent keep its memory between runs?
- Yes. In-memory filesystem snapshots keep a suspended sandbox ready to resume. For guaranteed long-term persistence, attach a Volume or use Agent Drive, which can retain data for years.
- How isolated is each agent?
- Every agent runs in its own hardware-isolated microVM with full access to its filesystem, processes, and logs, plus native support for Zero Data Retention.
- How long can an agent run?
- Sandboxes are built for workloads lasting seconds to hours. For longer background work you can pair them with Batch API, which runs up to 24 hours.
- What does it cost to keep an agent on standby?
- You stop paying for compute while a sandbox is suspended. You only pay the standby snapshot storage rate to keep it ready for instant resume.