# E2B

Official pricing: https://e2b.dev/pricing  
Category: agent-sandbox · Isolation: firecracker

## Pricing regimes (raw)

- **On-demand (per-second, allocated)** (resource): $0.0504/vCPU-h, $0.0162/GiB-h

## Features

Yes: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, volumes, idle auto-stop, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, full VM (own kernel), Docker inside, root, systemd, browser, desktop GUI, computer-use API, code interpreter, desktop control API, browser + desktop control, preinstalled agents, SSH, HTTPS preview URLs, egress allowlist, open internet, self-hosting, BYOC, open source, SOC 2, HIPAA, Python, Node.js, credential injection, wake on request, live fork (no pause), memory fork, webhooks, EU data residency, spend limits, MCP server, automatic snapshots, snapshots on demand, fast boot (benchmarked), gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), secret proxy, secret proxy for any API, extra volumes, shared volumes

No: nested virtualization, anti-bot stealth, CAPTCHA solving, residential IPs, public IPv4, custom domains, raw TCP inbound, static egress IP, SSO, GPU, arm64, Windows, macOS, GPU desktop, HTTP method/path egress rules, live resize, agent harness API, hosted agent API (their own agent)

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/e2b.json → feature_evidence

## Caveats

- CPU and RAM billed on allocation while running. Idle auto-pause is opt-in: default onTimeout is 'kill' (5 min default timeout); set lifecycle onTimeout 'pause' (+ autoResume) and the sandbox pauses after the timeout with no SDK/HTTP activity, then resumes on the next HTTP request or SDK call (docs.e2b.dev/sandbox/persistence, /sandbox/auto-resume, 2026-09-28). auto_stop_idle=true assumes you enable it.
- Pro $150/mo fee grants no usage credit.
- Enterprise: $3,000/mo minimum with 1-year commit; negotiated rates unpublished; treated here as a usage floor at list rates (unverified).
- BYOC (Enterprise, AWS/GCP) pricing unpublished; customer also pays own cloud compute.
- Egress/ingress: no network charges documented anywhere; left null (appears unmetered).
- Snapshots/paused sandboxes: officially not billed and retained indefinitely; some third-party blogs claim storage charges — not supported by official docs as of 2026-09-28.
- Volumes (private beta) and template build billing unpublished.
- Included disk 20 GiB is Pro ('20+'); Hobby is 10 GiB. No self-serve extra disk.
- Region multiplier for EU/Asia not published; assumed 1.
- Self-serve max 8 vCPU / 8 GiB; Pro '8+' on request. Allowed vCPU values: calculator lists 1,2,4,6,8; changelog says 1-8.
- No published minimum billed seconds or start fee; whether pause duration (~4 s per GiB RAM) is billed is undocumented.
- One-time $20,000 credit programmes (Startups, Research) exist for eligible first-time users; not modelled.
- Rates identical on Hobby and Pro; no volume discounts except via Enterprise.
- Verifier 2026-09-28: Pro+ ($650, 600 concurrent) and Pro++ ($1,150, 1,100 concurrent) were flagged 'addon', which the engine never prices; they are published self-serve plans on https://e2b.dev/pricing, so flags cleared.
- Sandbox creation-rate limits (Hobby 1/s, Pro 5/s; https://docs.e2b.dev/billing) are not modelled by the engine.
- Startups / Research programmes: one-time $20,000 credit plus Pro-tier membership ('The Pro tier, and $20,000 to spend on it'); concurrency add-ons and Enterprise bought separately. Startups: <3 years old, <$5M raised.
- BYOC (Enterprise) is now also available on Azure, in addition to AWS and GCP. Pricing is unpublished and the customer also pays their own cloud compute. Private cloud/on-prem is in development.
- Unresolved conflict: e2b.dev/pricing (2026-09-29) lists Hobby as 'Default sandbox CPU and RAM' (Pro: 'Custom'), while the docs/billing say Hobby allows up to 8 vCPU / 8 GiB. The card keeps 8/8 on Hobby.

## How this provider charges


E2B has one compute rate card (per-second, allocated vCPU + allocated RAM, identical on Hobby and Pro) but many
regimes that change what the same workload costs: plan fee that is *not* credit, concurrency add-ons, an Enterprise
usage minimum with a 1-year commit, one-time credit programmes, and lifecycle modes (running vs paused vs killed)
whose default is destructive. Storage, snapshots and egress are not billed per official docs.

Base rates: vCPU $0.000014/s = **$0.0504/vCPU-h**; RAM $0.0000045/GiB-s = **$0.0162/GiB-h**.
4 vCPU / 8 GiB = 0.2016 + 0.1296 = **$0.3312/h**.

## Regime table

| Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|
| Hobby (free) | Default for new accounts | No fee; usage paid from one-time credit; account **blocked** (not auto-charged) when credit runs out until a card is added | $0 fee; $100 one-time credit (~302 h of 4/8); 20 concurrent; 1 h max continuous run; 1 create/s; 8 vCPU / 8 GiB / 10 GiB disk max; 20 concurrent builds | https://e2b.dev/pricing, https://docs.e2b.dev/billing |
| Pro | Need >20 concurrent, >1 h continuous run, >10 GiB disk, EU/Asia region | $150/mo **pure fee — no credit** ("Upgrading to Pro does not grant additional credits") + per-second usage, billed monthly in arrears | $150/mo; 100 concurrent; 24 h max continuous; 5 creates/s; "8+" vCPU/GiB (higher via support); 20 GiB ("20+") disk | https://docs.e2b.dev/billing |
| Pro + concurrency add-on(s) | >100 concurrent self-serve | +$500/mo per add-on of 500 slots, on top of usage; monthly, no long-term commit; immediate | 600 concurrent = $650/mo total fixed; 1,100 = $1,150/mo total fixed (self-serve ceiling) | https://docs.e2b.dev/faq/increase-concurrency, pricing calculator |
| Enterprise (committed usage) | >1,100 concurrent, custom sizes, >24 h continuous, BYOC, "better rates" | Minimum monthly usage committed **for a year**; custom rates; custom billing | "$3,000/mo minimum" (pricing calculator); rates not published | https://e2b.dev/pricing, https://docs.e2b.dev/faq/increase-concurrency |
| BYOC | Enterprise only, AWS or GCP (Azure not yet) | E2B-managed deployment in customer VPC; customer pays own cloud compute; E2B fee unpublished | null | https://docs.e2b.dev/byoc |
| E2B for Startups | <$5M raised, founded <3 y, first-time E2B user; acceptance not guaranteed | One-time usage credit + Pro tier membership | $20,000 one-time credit | https://e2b.dev/startups |
| E2B for Research | University / public research / non-profit, first-time user | One-time usage credit + Pro tier membership | $20,000 one-time credit | https://e2b.dev/research |
| Running (active or idle) | Sandbox state = running | Allocated vCPU + allocated RAM per second, regardless of CPU/RAM utilisation; no published minimum or start fee | $0.0504/vCPU-h + $0.0162/GiB-h | https://docs.e2b.dev/faq/calculate-sandbox-price |
| Paused (memory + filesystem) | `pause()` or `onTimeout:'pause'` | Not billed; not counted toward concurrency; retained indefinitely (no TTL) | $0 | https://docs.e2b.dev/faq/paused-sandboxes-concurrency, https://docs.e2b.dev/faq/sandbox-lifetime |
| Paused filesystem-only (`keepMemory:false`) | Faster/lighter pause; reboot on resume | Not billed | $0 | https://docs.e2b.dev/sandbox/filesystem-only-snapshots |
| Named snapshots / forks | `createSnapshot`, `fork(count<=100)` | Snapshots: no billing documented; source sandbox is "briefly paused" during snapshot; each fork is a new running sandbox billed at full rate | $0 storage (not documented as billed) | https://docs.e2b.dev/sandbox/snapshots, https://docs.e2b.dev/sandbox/fork |
| Timeout (default) | No activity config; default timeout 5 min | Default `onTimeout` = **kill** (deletes state); billing stops | $0 after | https://docs.e2b.dev/faq/sandbox-lifetime |
| Disk | Per template, set by plan | Included; cannot buy more self-serve | 10 GiB Hobby / 20+ GiB Pro; "Storage: Free" | https://docs.e2b.dev/billing |
| Volumes | Private beta, US/EU | Price unpublished | null | https://docs.e2b.dev/volumes |
| Template builds | Building custom templates | Billing not documented; 1 h build timeout; 20 concurrent builds | null | https://docs.e2b.dev/faq/build-limits |
| Regions | us-west1 default; EU (Pro+ via support); Asia (Pro+ per features research) | No regional multiplier published; separate project/API key; templates rebuilt per region | multiplier assumed 1 (unverified) | https://docs.e2b.dev/faq/eu-region |
| Egress / ingress | All traffic | No network charge documented anywhere in docs or pricing page | null (appears unmetered) | https://docs.e2b.dev/network/internet-access |

Dated changes: billing enabled 2024-02-23; custom CPU/RAM (1–8 vCPU, 128 MiB–8 GiB) for Pro 2024-03-03;
self-serve concurrency add-on units in dashboard 2026-08-17. No rate change announced since. No upcoming
price change found (https://docs.e2b.dev/changelog).

## Gotchas

1. **Pro $150 is a pure fee.** It buys limits only; at low usage E2B costs $150 + usage, unlike boat/Daytona-style credit fees.
2. **Idle = full price.** CPU and RAM are billed on allocation while running. A sandbox left running and idle costs the same as one at 100% CPU. Pausing is the only way to stop the meter.
3. **Default timeout is destructive.** `onTimeout` defaults to `kill` after 5 min; people who expect "auto-stop" lose state. People who instead raise the timeout to 24 h to avoid the kill keep paying for idle hours. `Sandbox.connect()` never shortens and only extends the timeout (min 5 min from now), so reconnect loops can keep sandboxes alive.
4. **CPU/RAM are fixed per template, not per sandbox.** To change size you rebuild the template. Default base template is 2 vCPU / 512 MiB; a rebuild without parameters defaults to 2 vCPU / 1 GiB.
5. **Hobby is hard-stopped, not auto-charged**: when the $100 credit runs out the account is blocked. Pricing page says Hobby gets "Default sandbox CPU and RAM"; the billing docs say Hobby max 8 vCPU / 8 GiB (conflict — unresolved).
6. **Concurrency add-ons are steep and step-shaped**: going from 100 to 101 concurrent costs +$500/mo. Paused sandboxes don't count, so aggressive pausing is a cost lever.
7. **1 h continuous-run cap on Hobby / 24 h on Pro** resets on pause+resume; the "session cap" is not a lifetime cap.
8. **Enterprise needs $3,000/mo minimum and a 1-year commitment**; the minimum is described as a usage minimum (likely credited against usage, unverified).
9. **Third-party misinformation**: some blogs (Morph, Matrix OS) claim Pro "includes 500 hours" or that snapshot storage is billed. Official docs (2026-09) say Pro includes no credits and paused sandboxes/snapshots are not billed. "Template storage may be priced in future" was noted in earlier research, so watch for a change.
10. **No egress, IPv4, or volume prices are published.** Volumes are private beta with unpublished price; no static egress IP on any plan.
11. **Pause is not instant**: about 4 s per GiB RAM (8 GiB ≈ 32 s). Whether that pause window is billed is not documented.

## Worked example

Workload: 4 vCPU / 8 GiB, 50 concurrent × 8 h/day × 22 days = **8,800 sandbox-hours**, 30% CPU util,
50 GiB snapshots retained, 100 GiB egress. Sandboxes paused (not killed) between shifts.

Usage at list: 8,800 × $0.3312 = **$2,914.56** (vCPU $1,774.08 + RAM $1,140.48). CPU util 30% changes nothing (allocated billing).
Snapshots: $0 (paused state not billed). Egress: $0 (none documented, unverified).

| Regime | Feasible? | Monthly total |
|---|---|---|
| Hobby | No: 50 concurrent > 20; 8 h runs > 1 h cap (pause/resume every hour works around the cap but not concurrency) | n/a (the $100 credit covers ~0.3 h per sandbox-day of this load) |
| Pro on-demand | Yes (50 ≤ 100, 8 h ≤ 24 h) | **$3,064.56** = $150 + $2,914.56 |
| Pro + 600 add-on | Yes but unnecessary | $3,564.56 |
| Enterprise commit | Yes | ≥ **$3,000** (if the $3k minimum is a usage floor credited against usage and list rates apply; negotiated rates unknown; 1-year commit) |
| BYOC | Enterprise | unknown E2B fee + own AWS/GCP compute |
| Startup / Research credits | If accepted | $20k one-time covers ~6.9 months of usage; whether the $150 Pro fee is waived is not stated |
| Anti-pattern: left running idle 24 h/day instead of pausing | Yes | $150 + 50×24×22×0.3312 = **$8,893.68** |
| Anti-pattern: default onTimeout kill | — | Same $ as Pro but state is destroyed between shifts |