# Google Agent Platform Sandbox (Code Execution / Shell / Computer Use)

Official pricing: https://cloud.google.com/products/gemini-enterprise-agent-platform/pricing  
Category: agent-sandbox · Isolation: container

## Pricing regimes (raw)

- **Code Execution / Shell sandbox (API exec, same Agent Compute + Agent Memory meter)** (sizes): $0.085/vCPU-h, $0.009/GiB-h: default (MACHINE_CONFIG_UNSPECIFIED) 2 vCPU/1.5 GiB $0.1835/h; MACHINE_CONFIG_VCPU4_RAM4GIB 4 vCPU/4 GiB $0.376/h
- **Computer Use sandbox (managed browser, billed as Agent Compute + Agent Memory)** (resource): $0.085/vCPU-h, $0.009/GiB-h
- **Computer Use / Shell sandbox on Gemini Enterprise Flexible Savings Plan (1 year)** (resource): $0.0765/vCPU-h, $0.0081/GiB-h [commit]
- **Computer Use / Shell sandbox on Gemini Enterprise Flexible Savings Plan (3 year)** (resource): $0.068/vCPU-h, $0.0072/GiB-h [commit]

## Features

Yes: snapshots, memory snapshots, fork/clone, pause/resume, persistent disk, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, browser, computer-use API, code interpreter, browser automation API, desktop control API, browser + desktop control, egress allowlist, static egress IP, private networking, SOC 2, HIPAA, SSO, arm64, Python, Node.js, HTTP method/path egress rules, live fork (no pause), memory fork, EU data residency, snapshots on demand, agent harness API, inbound access rules

No: volumes, idle auto-stop, full VM (own kernel), Docker inside, nested virtualization, root, desktop GUI, anti-bot stealth, CAPTCHA solving, residential IPs, preinstalled agents, SSH, public IPv4, HTTPS preview URLs, open internet, self-hosting, BYOC, open source, GPU, Windows, macOS, wake on request, live resize, MCP server, automatic snapshots, hosted agent API (their own agent), gVisor or VM (no shared kernel), firewall inside (nftables), secret proxy, secret proxy for any API, extra volumes, shared volumes

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

## Caveats

- One meter, three sandbox types: Agent Compute $0.085/vCPU-h + Agent Memory $0.009/GiB-h (allocated, per-second) covers Agent Runtime, Code Execution, Shell and Computer Use sandboxes (Gemini Enterprise Agent Platform, formerly Vertex AI Agent Engine). The Runtime idle-between-turns exemption applies to Runtime only, not to sandboxes.
- None of these is an SSH-able VM: Computer Use is a containerized browser (API actions / CDP / VNC view); Code Execution is Python-only with no network access; Shell sandbox runs commands through an API. Sandbox vCPU/RAM sizes are not published.
- ENGINE_CARD cannot apply all independent free CPU/RAM/storage/request allowances exactly. Raw regimes/worked example retain them; any engine subtotal may overstate charges or miss eligibility. Do not subtract them twice.
- Features set to null were not verified. No sandbox was purchased or tested; feature/performance statements are documentation claims. Null maximums are unknown unless a cited note explicitly states unlimited.
- A complete monthly bill requires known snapshot/egress/plan limits; do not silently treat missing ancillary charges as free.

## How this provider charges


As of 2026-09-28. Prices are USD unless explicitly labelled otherwise. Source type: official published list/promo or unknown, never a fabricated quote.

Public sandbox pricing now appears under Gemini Enterprise Agent Platform (formerly Vertex AI Agent Engine). Agent Runtime idle-between-turns exception must not be generalized to sandbox CPU utilization. Sandbox compute billed on allocated resources.

| Regime | When it applies | How billed / numbers | Source |
|---|---|---|---|
| On-demand | Allocated sandbox resources | $0.085/vCPU-h + $0.009/GiB-h; nearest second. | https://cloud.google.com/products/gemini-enterprise-agent-platform/pricing |
| Commitment discount | Consumption model eligibility | Displayed alternative Agent Compute $0.0765/$0.068; corresponding memory and commitment selection not verified, excluded from numeric modes. | https://cloud.google.com/products/gemini-enterprise-agent-platform/pricing |

## Gotchas

- Public sandbox pricing now appears under Gemini Enterprise Agent Platform (formerly Vertex AI Agent Engine). Agent Runtime idle-between-turns exception must not be generalized to sandbox CPU utilization. Sandbox compute billed on allocated resources.
- ENGINE_CARD cannot apply all independent free CPU/RAM/storage/request allowances exactly. Raw regimes/worked example retain them; any engine subtotal may overstate charges or miss eligibility. Do not subtract them twice.
- Features set to null were not verified. No sandbox was purchased or tested; feature/performance statements are documentation claims. Null maximums are unknown unless a cited note explicitly states unlimited.
- A complete monthly bill requires known snapshot/egress/plan limits; do not silently treat missing ancillary charges as free.

## Worked workload

4 vCPU / 8 GiB; 50 simultaneous instances × 8 h/day × 22 days = **8,800 instance-hours**, **35,200 vCPU-hours**, **70,400 GiB-hours**. At 30% observed CPU:10,560 active-vCPU-hours only where the meter actually uses utilization. Assume one start/instance/day:1,100starts. Retain50GiB snapshots for a month;100GiB outbound. Active disk capacity was not specified.

Assuming a 4-vCPU/8-GiB template is supported and C=50 granted: 8,800×(4×.085+8×.009) = $3,625.60. Fully available monthly free pools remove 50×.085+100×.009=$5.15 → $3,620.45, before storage, egress and model calls. Shape/session/snapshot feasibility must be confirmed; this is not a guaranteed bill.

## Sources

- https://cloud.google.com/products/gemini-enterprise-agent-platform/pricing
- https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/sandbox/computer-use
- https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/sandbox/manage-sandboxes