# Moonshot Kimi Hosted Agents sandbox

Official pricing: https://platform.kimi.com/docs/pricing/hosted-agents.md  
Category: agent-sandbox · Isolation: not published

## Pricing regimes (raw)

- **Hosted Agents 1C/1G sandbox** (sizes): 1C1GB 1 vCPU/1 GiB $0.010729463537/h [alt, sales, beta]

## Features

Yes: pause/resume, persistent disk, volumes, idle auto-stop, ≥24 h sessions, browser, code interpreter, preinstalled agents, egress allowlist, open internet, Python, Node.js, Go, Rust, credential injection, wake on request, scheduled runs (cron), agent harness API, hosted agent API (their own agent), secret proxy, secret proxy for any API, extra volumes, shared volumes

No: snapshots, memory snapshots, fork/clone, start from your own snapshot, your own Docker/OCI image, Docker inside, nested virtualization, desktop GUI, computer-use API, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, SSH, public IPv4, HTTPS preview URLs, static egress IP, self-hosting, BYOC, open source, SOC 2, HIPAA, SSO, GPU, arm64, Windows, macOS, HTTP method/path egress rules, live resize, live fork (no pause), memory fork, EU data residency, zero data retention, MCP server, automatic snapshots, snapshots on demand, inbound access rules, firewall inside (nftables)

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

## Caveats

- 4/8 machine unavailable in published shape table; 50-way concurrency unknown.
- Tax basis not stated; do not assume CNY prices include 6% VAT.
- Pause of conversation is not proof of RAM/process snapshot persistence.
- 1 core may not equal dedicated physical core.
- Unknowns remain null, not free; generic infrastructure modes are opt-in alt.
- Worked example assumes resources are released between active windows unless a retained-instance scenario is explicitly discussed.
- VAT basis: Unknown: reviewed pricing source does not explicitly establish VAT/tax inclusion.
- Residency: China domestic enterprise-verified customers only, sales-assisted beta. Exact datacenter not published.
- Per-exec limits: shell/ipython single execution timeout 4 min, single tool call total timeout 5 min (defaults). Base image Debian 12. Each session has its own sandbox. Sandbox billed 'by running duration' -- whether an idle session (status idle, awaiting input) keeps a billed sandbox is not stated.

## How this provider charges


As of 2026-09-28. Price basis: public list/reference unless labeled historical, illustrative, promo or unpriced.

China-site Kimi Hosted Agents private enterprise beta has a real sandbox tariff, absent from the international .ai index. Published shape is only 1 CPU / 1 GB at CNY .072/hour; model tokens and plugins extra. Do not multiply to invent 4/8 availability. Sandbox state/conversation retained by service, so not covered by zero-data-retention.

## Currency, VAT and residency
- Native catalog: CNY. FX: 1 CNY = 0.149020326907 USD, ECB 2026-09-28 ([source](https://www.ecb.europa.eu/stats/policy_and_exchange_rates/euro_reference_exchange_rates/html/index.cs.html)). USD-listed catalogs remain independent quotes.
- Tax: Unknown: reviewed pricing source does not explicitly establish VAT/tax inclusion.
- Residency/restrictions: China domestic enterprise-verified customers only, sales-assisted beta. Exact datacenter not published.

## Regimes
| Regime | When it applies | How billed | Native / USD numbers | Source |
|---|---|---|---|---|
| hosted-beta | Hosted Agents 1C/1G sandbox | allocated wall time; 1 second increment | 1C1GB 1/1: CNY 0.072/h → USD 0.010729463537; month cap None | https://platform.kimi.com/docs/pricing/hosted-agents.md |
| model | Inference | See rule | Chosen model tokens charged separately. | https://platform.kimi.com/docs/pricing/hosted-agents.md |
| plugins | Optional tools | See rule | CNY 1/20/30/45/60/110/150/450 per 1000 calls by plugin. Search 30, image search 45. | https://platform.kimi.com/docs/pricing/hosted-agents.md |

## Gotchas
- 4/8 machine unavailable in published shape table; 50-way concurrency unknown.
- Tax basis not stated; do not assume CNY prices include 6% VAT.
- Pause of conversation is not proof of RAM/process snapshot persistence.
- 1 core may not equal dedicated physical core.
- Unknowns remain null, not free; generic infrastructure modes are opt-in alt.
- Worked example assumes resources are released between active windows unless a retained-instance scenario is explicitly discussed.
- VAT basis: Unknown: reviewed pricing source does not explicitly establish VAT/tax inclusion.
- Residency: China domestic enterprise-verified customers only, sales-assisted beta. Exact datacenter not published.

## Required worked example
4 vCPU / 8 GiB requested; 50 simultaneous workers × 8 hours/day × 22 days = **8,800 instance-hours**, 1,100 daily sessions. CPU utilization 30% does not reduce allocated-resource charges. Snapshot retention is **50 GiB total for 730 hours**, not 50 GiB per worker; egress is **100 GiB total**. Root disk size was not specified, so root/volume costs are not silently assigned zero.
- hosted-beta: compute **unknown**; cannot produce a comparable quote.
- 50 GiB retained snapshots: **unknown/unverified feature or price**, not USD 0.
- 100 GiB egress: **unknown**, not USD 0.
- Full all-in bill: **null** where any required storage/network/fee or eligibility fact is unresolved. Quota raises may be necessary for 200 aggregate vCPU / 400 GiB RAM.

## Sources
- https://platform.kimi.com/docs/pricing/hosted-agents.md
- https://platform.kimi.com/docs/hosted-agents/quickstart.md
- https://platform.kimi.com/docs/hosted-agents/sandbox-reference.md
- https://platform.kimi.com/docs/hosted-agents/environments.md