boat vs Prime Intellect
Both give each agent its own microVM with its own kernel. Prime Intellect is built to run thousands of short RL and eval rollouts at once, while boat gives an agent a machine it can stop, come back to and host from, with agents and a desktop already inside.
Where boat stands out
A machine you can stop and come back to
A boat sandbox keeps its disk when it stops, takes a snapshot every minute, and costs nothing while stopped. You start the same machine again later with every file in place, and you can fork a running one.
Agents you drive from outside, with keys kept out
Claude Code, Codex and five other agents come preinstalled, and you prompt, steer and follow them through one API. Real API keys stay in a credentials proxy outside the sandbox, and a credential can switch to another account when one runs out of credits.
Hosting and a desktop in the box
Each sandbox gets HTTPS URLs on up to 50 ports, raw TCP, and a 60fps Linux desktop with Chromium for computer use. It also exposes /dev/kvm for Android emulators and VMs.
The companies and products
boat
boat.dev runs sandboxes for AI agents: each one is a full Linux VM with its own kernel.
You start a machine with one command or one API call, then SSH in, run Docker, open a desktop, or let a built-in agent (Claude Code, Codex, OpenCode and others) work in it. The disk stays between sessions, and a stopped sandbox costs nothing.
- Isolation
- Full VM per sandbox, own kernel
- Regions
- EU: Germany, Finland, France
Prime Intellect
Prime Intellect sells an open AI stack for training and running your own models: GPU compute, RL training, evaluations, inference and sandboxes.
Prime Sandboxes are disposable microVMs that boot a Docker image you choose, driven from a Python SDK or the prime CLI. They are made for RL research and evals, plug into Prime's open-source verifiers and prime-rl libraries, and are built to start thousands of sandboxes at once. A sandbox lives until its timeout or idle timer ends it, and you can save its files as a checkpoint to start new sandboxes from.
Which use cases each one fits
For each use case, Battleships prices every sandbox provider on the same workload and sorts them from the cheapest monthly bill up. “3rd cheapest of 367” means 2 of the other 366 providers run that workload, with every feature it needs, for less. “With limits” means the provider runs it but misses something, so it is listed after every full match. Open a use case for its whole ranking and what each provider misses.
| Use case and its workload | boat: price rank | Prime Intellect: price rank |
|---|---|---|
| Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month | 3rd cheapest of 367 $634 a month | 193rd of 367, with limits $2,243 a month |
| Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB An assistant that lives on its machine all month needs a disk that survives stops, which boat has; on Prime the sandbox must keep running or be rebuilt from a checkpoint. | 5th cheapest of 366 $657 a month | 251st of 366, with limits $3,431 a month |
| Code interpreter bursts 100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month Prime Intellect fits short code runs well: it starts thousands of sandboxes in seconds and bills only while each one runs. | 186th of 345, with limits $2,000 a month | 15th cheapest of 345 $151 a month |
| Sandboxes for your users 300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month boat gives each user HTTPS on up to 50 ports, raw TCP and a disk that persists, so a user's app can be hosted from the sandbox. | Cheapest of 361 $209 a month | 13th cheapest of 361 $957 a month |
| RL rollouts 1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month RL is what Prime Sandboxes are made for, with verifiers and prime-rl built in, while boat's start rate limits how fast a big batch can begin. | 34th of 360, with limits $2,000 a month | 82nd of 360, with limits $4,574 a month |
| Agent evals 250 at once · 4 vCPU / 8 GB · 15,000 sessions of 20 min a month Both fit; Prime also offers hosted evaluations, and boat gives each eval a faster-to-debug machine with agents preinstalled. | 10th cheapest of 364 $200 a month | 41st cheapest of 364 $736 a month |
| CI runners 40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month | 4th cheapest of 377 $86 a month | 60th cheapest of 377 $451 a month |
| 24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB A dev box needs to stop at night and come back with its files, which boat does on its own and Prime does only through manual checkpoints. | 18th cheapest of 368 $26 a month | 96th cheapest of 368 $139 a month |
| Computer-use agents 50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month Computer-use agents need a desktop and a browser in the machine, which boat streams at 60fps and Prime Sandboxes do not provide. | 6th cheapest of 369 $200 a month | 202nd of 369, with limits $662 a month |
| Browser-use agents 50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month Prime's browser environments call Browserbase, a separate service, while boat runs Chromium inside the sandbox. | 7th cheapest of 367 $200 a month | 233rd of 367, with limits $331 a month |
| Scraping & bots (startup scale) 50 at once · 1 vCPU / 2 GB · 100,000 sessions of 3 min a month | 6th cheapest of 346 $90 a month | 235th of 346, with limits $224 a month |
| Mass scraping & bots (undetectable) 1,000 at once · 1 vCPU / 2 GB · 3,000,000 sessions of 2 min a month | 71st of 338, with limits $1,800 a month | 152nd of 338, with limits $4,485 a month |
| Parallel red teaming 200 at once · 2 vCPU / 4 GB · 5,000 sessions of 30 min a month | 2nd cheapest of 360 $200 a month | 212th of 360, with limits $225 a month |
| GPU eval jobs 10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month | Can't run it | 63rd of 98, with limits $828 a month |
| macOS agents 5 at once · 4 vCPU / 8 GB · 1,000 sessions of 20 min a month | Can't run it | Can't run it |
Key features
| Feature | boat | Prime Intellect |
|---|---|---|
| Isolation | Full VM with its own kernel on dedicated AMD Ryzen 9 9950X hosts; root, Docker and systemd | Hardware-isolated microVM with its own kernel; Docker, Docker Compose and systemd |
| Nested virtualization | Yes, /dev/kvm for Android emulators and VMs | Not documented |
| Biggest machine | Up to 16 vCPU / 32 GB on the $100 plan and above | Up to 16 vCPU / 64 GiB RAM and 128 GiB disk |
| Session length | No cap on paid plans; you set the time to live | 60 minutes by default; can be set to no deadline |
| Stop and resume | Stop is free; the disk is kept and the same sandbox starts again | No stop; timeouts and the idle timer delete the sandbox and its disk |
| Saving state | Disk snapshot every minute and on stop, automatically | Manual filesystem checkpoints at $0.10 per GiB a month, restored into a new sandbox |
| Fork | Fork a running sandbox with its disk | Fork a running sandbox's files; processes start fresh |
| Start many at once | Up to 90 starts a minute on the $2,000 plan; faster bursts are queued | Built to start thousands of sandboxes within seconds |
| Inbound access | HTTPS URLs for up to 50 ports, token-protected by default; raw TCP | No port exposure on sandboxes; Prime Tunnel gives an HTTPS URL, open by default or behind basic auth, documented inside sandboxes only for hosted evals; no raw TCP |
| SSH | Full SSH and SCP | CLI SSH with a one-time key; no port forwarding, scp or sftp |
| Outbound rules | Open internet, no provider-side allowlist | Allow or deny lists by domain or IPv4 CIDR, changeable while running |
| Coding agents | Seven agents preinstalled, with a prompt, steer and events API you drive from outside | verifiers library runs Claude Code, Codex and other harnesses to completion for evals and RL |
| Keeping API keys out of the sandbox | Credentials proxy for model providers and any HTTPS API; switch a credential to another account when credits run out | Secrets injected as environment variables inside the sandbox |
| Desktop and browser | 60fps Linux desktop with Chromium and computer use | Headless; browser environments use Browserbase, a separate service |
| Compliance and transparency | SOC 2 is in progress. We're open about where we stand: live status and data retention policy. | SOC 2 |
Every feature Battleships tracks (43)
| Feature | boat | Prime Intellect |
|---|---|---|
| Full VM (own kernel) | Yes | Yes |
| Docker inside | Yes | Yes |
| Nested virtualization | Yes | No |
| Root | Yes | Not published |
| Systemd | Yes | Yes |
| Desktop GUI | Yes | No |
| Browser | Yes | No |
| Computer-use API | Yes | No |
| SSH | Yes | Yes |
| Persistent disk | Yes | No |
| Pause/resume | Yes | No |
| Automatic snapshots | Yes | No |
| Fork/clone | Yes | Yes |
| Live fork (no pause) | Yes | Yes |
| Memory fork | No | No |
| Custom image (Docker or snapshot) | Yes | Yes |
| ≥24 h sessions | Yes | Yes |
| Idle auto-stop | No | Yes |
| Wake on request | No | No |
| HTTPS preview URLs | Yes | No |
| Raw TCP inbound | Yes | No |
| Public IPv4 | No | No |
| Egress allowlist | No | Yes |
| Secret proxy | Yes | No |
| Secret proxy for any API | Yes | No |
| Preinstalled agents | Yes | No |
| Agent harness API | Yes | Yes |
| MCP server | No | No |
| GPU | No | Yes |
| Windows | No | No |
| Arm64 | No | No |
| Self-hosting | No | No |
| BYOC | No | No |
| Open source | No | No |
| Scoped API keys | Yes | Not published |
| Spend limits | Yes | Yes |
| Usage per sandbox (API) | Yes | No |
| Audit logs | No | Not published |
| SSO | No | No |
| SOC 2 | No | Yes |
| HIPAA | No | No |
| EU data residency | Yes | Yes |
| Zero data retention | Yes | Not published |
Performance
There is no comparable measurement: Prime Sandboxes was not in the hpc-sandbox-benchmarks run. On that benchmark's Node.js build-and-test loop at 4 vCPU / 8 GB, boat did 26.69 runs per second.
Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. Prime Intellect has no published run in this suite yet, so there is no like-for-like number to compare.
Pricing in five scenarios
Each bill is priced from the providers' public price lists by the open Battleships engine, with the cheapest plan that runs the scenario.
| Scenario | boat, a month | Prime Intellect, a month |
|---|---|---|
| One coding agent, workdays 4 vCPU / 8 GB, 50 GB disk, 8 hours a day, 22 days a month, files kept between days | $20 · $20/mo | $30 · Pay-as-you-go · with limits |
| One agent that never stops 4 vCPU / 8 GB, 50 GB disk, running 24/7 all month | $26 · $20/mo | $139 · Pay-as-you-go · with limits |
| 50 agents in parallel 50 machines of 4 vCPU / 8 GB at once, 8-hour sessions, 1,100 sessions a month, Docker inside | $317 · $20/mo × 3 seats | $1,457 · Pay-as-you-go · with limits |
| Sandboxes for your users 2 vCPU / 4 GB, 30,000 sessions of 10 minutes a month, up to 300 at once | $200 · $100/mo × 2 seats | $414 · Pay-as-you-go |
| 1,000 short jobs at once 2 vCPU / 4 GB, 1,000 at once, 3 minutes each, 200,000 jobs a month | $2,000 · $2000/mo · with limits | $915 · Pay-as-you-go |
boat: Plans start at $20 a month, and the whole fee is usage credit. You pay per second while a sandbox runs, and a stopped sandbox costs nothing and keeps its disk.
Prime Intellect: Prime bills the vCPU, RAM and disk you allocate while the sandbox runs, at launch rates valid through December 22, 2026, from a prepaid wallet with no subscription. Checkpoints cost $0.10 per GiB a month, and later rates are not published.
Limits in those scenarios
Prime has no way to stop a sandbox and keep its disk, so the workday agent, always-on assistant and agent fleet scenarios, where an agent keeps its files between sessions, are marked as not fitting: you either keep it running and paying, or rebuild it from a manual checkpoint. Prime's default account caps of 4,096 vCPU and 4,096 GiB RAM allow about 512 sandboxes at 4 vCPU / 8 GiB. boat keeps every disk, but its start limits are lower: for 1,000 short jobs at once, even the $2,000 plan queues starts, while Prime starts them all.
| Scenario | boat | Prime Intellect |
|---|---|---|
| One coding agent, workdays | None that bite |
|
| One agent that never stops | None that bite |
|
| 50 agents in parallel | None that bite |
|
| Sandboxes for your users | None that bite |
|
| 1,000 short jobs at once |
|
|
Which one should you use?
Use Prime Intellect when
you run large batches of short RL or eval rollouts that start together and throw their state away, or want sandboxes next to the GPUs you train on.
- Built for RL at scale. For RL and eval runs: verifiers and prime-rl built in, and thousands of sandboxes started within seconds.
- Outbound allow and deny lists. Useful when an agent should only reach the hosts you pick, with lists you can change while it runs.
- More memory and a GPU cloud next door. A sandbox can have up to 64 GiB of RAM, and the same account rents GPUs for training.
Use boat when
your agents work for hours or days on a machine that keeps its files, hosts what they build, runs a desktop and a browser, and keeps your API keys out of reach.
- A machine you can stop and come back to. On Prime the timeout or idle timer deletes the sandbox and its disk. Keeping work means checkpointing by hand, paying $0.10 per GiB a month, and booting a fresh sandbox.
- Agents you drive from outside, with keys kept out. Prime's harnesses run an agent to completion for an eval, and its secrets land as environment variables inside the VM, where a prompt injection can read and leak them.
- Hosting and a desktop in the box. Prime Sandboxes expose no ports: its tunnel is documented inside a sandbox only for hosted evals and is open by default, raw TCP is not offered, and a browser agent has to call Browserbase, a separate paid service.
Questions people ask
Is boat cheaper than Prime Intellect?
In four of the five scenarios on this page, yes. For 1,000 short jobs at once, Prime starts them all for $915 a month while boat's $2,000 plan still queues the starts, so Prime fits that one better.
Does a Prime Intellect sandbox keep its files when it stops?
No. When its timeout or idle timer ends, the sandbox and its disk are deleted. You can save the files first as a checkpoint and start a new sandbox from it, but running processes are not kept. boat keeps the disk when a sandbox stops and starts the same machine again.
Can I run Claude Code in each?
Yes. Prime's verifiers library runs Claude Code, Codex and other harnesses inside sandboxes for eval and RL runs. boat ships seven agents preinstalled, with an API to prompt, steer and follow them while they work.
Can I run Docker in each?
Yes. Both are microVMs with their own kernel that run Docker and systemd. boat also exposes /dev/kvm for Android emulators and VMs, which Prime's docs do not mention.
Does Prime Intellect have GPU sandboxes?
Not yet. Prime's docs say GPU sandboxes are coming soon and need a grant. Its GPU Pods, a separate product, rent GPUs today.
Is boat SOC 2 compliant?
SOC 2 is in progress. We're open about where we stand: live status at status.ascii.dev and our data retention policy at docs.boat.dev/data-retention.
Start a sandbox
Install the CLI, then run boat new.
7-day free trial