# Hopx

Official pricing: https://hopx.ai/pricing  
Category: agent-sandbox · Isolation: firecracker

## Pricing regimes (raw)

- **Pay-per-use (per second, allocated, template-fixed size)** (resource): $0.0504/vCPU-h, $0.0162/GiB-h

## Features

Yes: snapshots, memory snapshots, pause/resume, persistent disk, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, full VM (own kernel), browser, desktop GUI, computer-use API, code interpreter, browser + desktop control, HTTPS preview URLs, egress allowlist, open internet, SOC 2, SSO, Python, Node.js, Go, webhooks, EU data residency, MCP server, snapshots on demand, gVisor or VM (no shared kernel), firewall inside (nftables)

No: fork/clone, volumes, idle auto-stop, Docker inside, nested virtualization, anti-bot stealth, CAPTCHA solving, residential IPs, preinstalled agents, SSH, public IPv4, raw TCP inbound, static egress IP, self-hosting, BYOC, open source, HIPAA, GPU, arm64, Windows, macOS, credential injection, HTTP method/path egress rules, wake on request, live resize, live fork (no pause), memory fork, automatic snapshots, agent harness API, hosted agent API (their own agent), fast boot (benchmarked), inbound access rules, secret proxy, secret proxy for any API, extra volumes, shared volumes

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

## Caveats

- CPU billing basis is not stated. The rates are labelled per allocation and the page examples compute on allocated time, so this card treats CPU as alloc. 'Pay only when your code runs' refers to paused/deleted sandboxes, not idle-running ones.
- The pricing-page example '1 vCPU/1 GB/10 GB for 1 h = $0.05' is inconsistent with the published rates, which give $0.0677.
- Paused/stopped storage, snapshot storage and egress prices are unpublished (null). The pricing page claims 'no idle charges' for paused sandboxes.
- No idle auto-stop. The timeout (default 1 h, extendable) deletes the sandbox and its data. Idle-running sandboxes bill full allocation until paused, killed or deleted by the timer.
- Sizes are template-fixed (set at build time). Per-plan vCPU/RAM caps and concurrency caps are unpublished. A CLI sample shows Free = 10 sandboxes, Pro = 100 (unverified).
- Pro plan price is unpublished on the pricing page ($29/mo appears only in illustrative CLI output). Enterprise volume discounts are unpublished.
- The 'gpu' template feature flag exists, but no GPU types or prices are documented.
- Benchmarks: computesdk cold-start success rate is 62% (median 4.2 s), burst success 6% (median 69 s), and DAX never succeeded. Latest runs are 0% successful. No hpc CPU benchmark.
- Included disk is assumed 0. Disk is billed per GiB-second on the template's allocated disk while running.
- unverified: may bill active CPU. hopx.ai/pricing markets "Pay Only When Your Code Runs" / "No idle charges"; docs and examples indicate allocated resources while running (idle = paused). alloc kept.
- Re-checked 2026-09-28: hopx.ai/pricing lists only vCPU, RAM and storage per-second rates; no snapshot/paused-storage or egress price is published (snapshot_gib_month and egress_gib stay null). Browser: hopx.ai advertises "Optional browser automation (Firefox/Chromium)" with VNC/noVNC in the desktop template, so features.browser=true.
- The $200 sign-up credit is one-time and 'may expire under posted program rules' (no expiry date published).
- The hopx.ai site footer shows a static 'All systems operational' badge even while status.hopx.ai shows the Public API down. Rely on status.hopx.ai.
- Activity signals: last product-update blog post is 2025-11-25 (MCP server); GitHub hopx-ai/hopx last push 2026-07-09, cli 2026-05-31. The sandbox heartbeat monitor on the status page is paused.
- Availability warning (2026-09-29): the status page reported the HopX Public API down since about 2026-09-27 07:05 UTC, and api.hopx.dev returned Cloudflare 526 (invalid SSL). No shutdown was announced. Re-check before relying on this provider.

## How this provider charges


Hopx (by Bunnyshell, Firecracker microVMs) has **one public price sheet** and no subscription: $0.000014/vCPU-s, $0.0000045/GiB-s RAM, $0.00000003/GiB-s disk, per second. The rates are byte-for-byte identical to E2B's. A 4 vCPU / 8 GiB sandbox costs **$0.3312/h** plus disk. Beyond that, what changes the bill is the lifecycle state (running / paused / stopped / timed-out), the wall-clock timeout, and the unpriced Pro/Enterprise tiers.

## Regime table

| # | Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|---|
| 1 | Pay-per-use (default, "Free tier") | Every account. No subscription, no minimum | Per second on **allocated** vCPU + RAM + disk while running. The page labels memory "RAM allocation". The "100 × 30 s × 1 vCPU = $0.04" example also only works as allocated wall-clock (3,000 s × $0.000014 = $0.042) | $0.0504/vCPU-h, $0.0162/GiB-h, $0.000108/GiB-h disk ($0.0788/GiB-month). 4/8 = $0.3312/h + $0.00108/h per 10 GiB disk | https://hopx.ai/pricing |
| 2 | "Pay only when your code runs" marketing reading | If CPU were metered on utilisation | **Not supported by the rate labels.** The page's own examples compute on allocation. Kept only as a sensitivity row | would be 4 × util × 0.0504 + 8 × 0.0162 | https://hopx.ai/pricing |
| 3 | Custom sizes (template-fixed) | Resources are set per **template** at build time, not at sandbox create | Same linear rates at any size | 1-64 vCPU, 512 MB-64,000 MB RAM, 1-250 GB disk. Limits "depend on your plan" (per-plan values unpublished) | https://docs.hopx.ai/core-concepts/templates/configuration |
| 4 | Desktop / computer-use templates | VNC desktop, screen recording, mouse/keyboard API | No separate desktop surcharge published. Same resource rates | = regime 1 | https://docs.hopx.ai/core-concepts/desktop/index |
| 5 | Running but idle | Between calls, until the timeout fires | Full allocated rate. There is **no idle auto-stop**: `timeout_seconds` is a wall-clock auto-**delete** timer | Default timeout 3,600 s (CLI config). Extendable via PUT /timeout | https://docs.hopx.ai/core-concepts/sandboxes/timeout |
| 6 | Paused | `POST /v1/sandboxes/:id/pause` (memory + processes preserved) | Compute stops ("No idle charges", "Pause to free resources"). **Storage/snapshot charge while paused not published** | compute $0; storage null | https://hopx.ai/pricing, https://docs.hopx.ai/api/control-plane/pause-sandbox |
| 7 | Stopped | `stop` (filesystem kept, full restart on start) | Compute stops. Whether disk ($0.0788/GiB-month) keeps billing is **not documented** | null | https://docs.hopx.ai/core-concepts/sandboxes/managing-state |
| 8 | Timed out | Timeout expires | Sandbox is **deleted**, all data lost, billing ends | $0 | https://docs.hopx.ai/core-concepts/sandboxes/timeout |
| 9 | Snapshots | `hopx system snapshot` CLI id; template snapshots | No published price. The provider record notes a 30-days-after-last-use expiry (configurable). No documented restore/fork flow | null | https://docs.hopx.ai/llms-full.txt |
| 10 | Regions | `region` = `us-east` or `eu-west` at create | **No regional price difference published** | multiplier 1 | https://docs.hopx.ai/api/control-plane/create-sandbox |
| 11 | Egress / public URLs | Every port auto-exposed at `https://<port>-<id>.hopx.dev` | Egress **not priced on the page** (neither free nor metered). No public IPv4 | null | https://hopx.ai/pricing |
| 12 | Free credit | Sign-up, no card | $200 one-time. "Enough for ~4,000 hours of basic usage". Expiry not stated | $200 | https://hopx.ai/pricing |
| 13 | Prepaid balance + auto-recharge | Dashboard/CLI `hopx billing auto-recharge` | Usage deducts from a balance. Recharge amounts are not documented | null | https://docs.hopx.ai/cli/commands/billing |
| 14 | Pro tier | Referenced in rate-limit docs ("2x default limits") | **Price not on the pricing page.** The only figure is an illustrative CLI `hopx usage plans` output: "Free $0 10 sandboxes / Pro $29/mo 100 sandboxes / Enterprise Custom". Treated as unverified (fee null) | 2× rate limits; $29/mo and 100 sandboxes only in sample output | https://docs.hopx.ai/api/concepts/rate-limits, https://docs.hopx.ai/cli/commands/usage |
| 15 | Enterprise | Contact sales | Custom SLAs, dedicated support, **volume discounts**. Unpublished | null | https://hopx.ai/pricing |
| 16 | Rate limits (all plans) | Default | 20 sandbox creates/min, 100 control-plane req/min, 300 VM-agent req/min, 10 template builds/h | Pro 2×, Enterprise custom | https://docs.hopx.ai/api/concepts/rate-limits |

## Gotchas

1. **Same list rates as E2B, but less clarity.** Hopx does not publish plan concurrency caps, paused/stopped storage charges, snapshot prices or egress prices.
2. **The pricing page's own example is wrong.** "1 vCPU, 1 GB, 10 GB storage, 1 hour = $0.05" computes to $0.0504 + $0.0162 + $0.00108 = **$0.0677**. "8 h, 2 vCPU, 4 GB ≈ $1.50" computes to $1.32 plus disk.
3. **The timeout is a delete, not a pause.** `timeout_seconds` is wall-clock. When it fires the sandbox is destroyed with its data. Nothing auto-pauses on idle, so an idle sandbox bills the full allocation until you pause/kill it or the timer deletes it.
4. **"No idle charges" means paused, not idle-running.** A running sandbox that is doing nothing still pays 100% of allocated vCPU/RAM.
5. **Sizes live in the template.** You can't request 4/8 at create. Build a template with those resources. Per-plan max vCPU is "from your plan", which is unpublished.
6. **Pro plan is a ghost.** It appears in the rate-limit docs and a CLI sample ($29/mo), but you can't buy it from the pricing page.
7. **Reliability.** ComputeSDK benchmarks: cold-start success rate 62%, burst (100 concurrent) success 6%, DAX never succeeded. The last recorded runs (Aug/Sep 2026) had a 0% success rate. Budget for retries, and any retries bill.

## 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.

Assumptions:
- The template disk is 10 GiB, billed while running: 8,800 h × 10 × $0.000108 = **$9.50**.
- The 50 concurrent sandboxes are within the 20 creates/min rate limit (3 min to spin up the fleet).
- The plan's concurrency cap is unknown.

| Regime | Compute | Disk running | Paused/stopped storage | Snapshots 50 GiB | Egress 100 GiB | Plan fee | **Monthly total** |
|---|---|---|---|---|---|---|---|
| Pay-per-use, sandboxes killed after each day | 8,800 × 0.3312 = $2,914.56 | $9.50 | $0 | null (unpublished) | null (unpublished) | $0 | **$2,924.06** + snapshots + egress |
| Pay-per-use, paused overnight (554 h × 50) | $2,914.56 | $9.50 | null (unpublished) | null | null | $0 | **$2,924.06** + paused storage + snapshots + egress |
| Idle until the default 1 h timeout instead of explicit kill | +1,100 h × 0.3312 = +$364.32 | +$1.19 | | | | | **$3,289.57** + unknowns |
| Hypothetical active-CPU reading (not supported by the rates) | 8,800 × (4 × 0.3 × 0.0504 + 8 × 0.0162) = $1,672.70 | $9.50 | | | | | **$1,682.20** (illustrative only) |
| Pro (unpriced) / Enterprise (volume discounts) | unpublished | | | | | null | null |

- **30% CPU utilisation does not reduce the bill.** Billing is on allocation.
- The $200 one-time credit covers month 1 only (≈ $16.67/month over 12 months).
- If snapshots/paused state were billed at the disk rate ($0.0788/GiB-month), 50 GiB would cost $3.94. That rate is illustrative only.

Sources: https://hopx.ai/pricing · https://docs.hopx.ai/llms-full.txt · https://docs.hopx.ai/api/concepts/rate-limits · https://docs.hopx.ai/cli/commands/billing · https://docs.hopx.ai/cli/commands/usage · https://docs.hopx.ai/core-concepts/templates/configuration · https://docs.hopx.ai/core-concepts/sandboxes/timeout · https://docs.hopx.ai/core-concepts/sandboxes/managing-state · https://docs.hopx.ai/api/control-plane/pause-sandbox · https://www.computesdk.com/benchmarks/sandboxes/