boat vs OpenReward
OpenReward is a platform for training and evaluating agents on reinforcement learning environments, and its sandboxes are the containers those environments run code in. boat gives an agent a whole VM of its own that keeps its disk, runs Docker and can be reached from outside.
Where boat stands out
A whole VM, not a container
Each boat sandbox is a full VM with its own kernel, root, systemd and Docker inside, on dedicated AMD Ryzen 9 9950X hosts. It also exposes /dev/kvm, so Android emulators and VMs start, and it scales up to 16 vCPU / 32 GB.
A disk that stays, and forks
boat saves the disk every minute and when the sandbox stops, and a stopped sandbox costs nothing. You can fork a running sandbox with its disk to try two paths from the same state, and there is no session cap on paid plans.
Reach it from outside
boat gives each sandbox SSH, HTTPS URLs on up to 50 ports, raw TCP and its own IP, plus a 60fps desktop with Chromium for computer use. Coding agents come preinstalled, and you drive agents inside from outside with a prompt, steer and events API.
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
OpenReward
OpenReward, built by GR Inc, is a platform to train and evaluate agents on 380+ reinforcement learning environments through one API.
You sample rollouts from hosted environments and use their verified rewards in your training loop. When a task needs a computer, the environment starts an OpenReward Sandbox: a container you run commands in and move files to and from, deleted when the session ends. OpenReward's own docs say the sandbox is not a general sandbox compute solution and is meant to be tied to environment use, so this page compares the two where they overlap: running agent code in an isolated Linux machine.
- Isolation
- Container per sandbox
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 | OpenReward: price rank |
|---|---|---|
| Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month A software factory needs Docker, a disk that survives a stop and a URL to preview the app, none of which an OpenReward sandbox offers. | 3rd cheapest of 367 $634 a month | 260th of 367, with limits $3,993 a month |
| Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB An assistant that lives on its machine needs files that persist; an OpenReward sandbox is deleted when its session ends. | 5th cheapest of 366 $657 a month | 291st of 366, with limits $6,044 a month |
| Code interpreter bursts 100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month Short code runs suit OpenReward's per-second containers, which start, run a command and are deleted. | 186th of 345, with limits $2,000 a month | 130th of 345, with limits $271 a month |
| Sandboxes for your users 300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month | Cheapest of 361 $209 a month | 273rd of 361, with limits $1,704 a month |
| RL rollouts 1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month OpenReward is built around RL: it hosts the environments and rewards, and starts a sandbox per session; boat gives each rollout a full VM with Docker and a disk you can fork. | 34th of 360, with limits $2,000 a month | 127th of 360, with limits $8,234 a month |
| Agent evals 250 at once · 4 vCPU / 8 GB · 15,000 sessions of 20 min a month OpenReward runs coding agents against its catalogue of benchmark environments; boat gives each eval its own VM with the agent preinstalled and Docker for real test suites. | 10th cheapest of 364 $200 a month | 163rd of 364, with limits $1,325 a month |
| CI runners 40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month CI runners usually build Docker images or start emulators, which needs Docker inside and /dev/kvm, both on boat. | 4th cheapest of 377 $86 a month | 319th of 377, with limits $1,106 a month |
| 24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB A dev box needs SSH and a disk that stays, which boat has and OpenReward's sandboxes do not. | 18th cheapest of 368 $26 a month | 313th of 368, with limits $242 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; boat streams a 60fps Linux desktop, and OpenReward's sandboxes are headless. | 6th cheapest of 369 $200 a month | 255th of 369, with limits $1,192 a month |
| Browser-use agents 50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month | 7th cheapest of 367 $200 a month | 265th of 367, with limits $596 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 | 269th of 346, with limits $404 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 | 204th of 338, with limits $8,073 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 | 256th of 360, with limits $393 a month |
| GPU eval jobs 10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month | Can't run it | Can't run it |
| 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 | OpenReward |
|---|---|---|
| Kind of product | Cloud VMs for agents, used for any workload | Sandboxes inside an RL environment platform, not meant for general compute |
| Isolation | Full VM with its own kernel on dedicated AMD Ryzen 9 9950X hosts | Container |
| Docker and nested virtualization | Docker, systemd and root inside; /dev/kvm for Android emulators and VMs | No Docker inside or /dev/kvm documented |
| Biggest machine | Up to 16 vCPU / 32 GB on the $100 plan and above | Up to 4 vCPU / 16 GiB, or one NVIDIA L4 GPU preset |
| Disk after a stop | Kept, with a snapshot every minute; a stopped sandbox costs nothing | Deleted: stop() terminates the sandbox; buckets mount read-only |
| Fork | Fork a running sandbox with its disk | No fork; start a new sandbox from the same image |
| Access | SSH, HTTPS URLs for up to 50 ports, raw TCP and its own IP | run, upload and download through the SDK; no SSH or inbound URL |
| Desktop and browser | 60fps Linux desktop with Chromium and computer use | Headless; no desktop or browser |
| Coding agents | Seven agents preinstalled in the sandbox, with a prompt, steer and events API | Firehorse runs Claude Code, Codex, Gemini and two API agents, with file tools proxied into the sandbox over MCP |
| 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 replaced by placeholders and swapped back on outbound HTTP only to the hosts set for each secret |
| Outbound network | Open internet, no provider-side switch | Open by default; block_network=True cuts all outbound traffic |
| GPUs | No | NVIDIA L4 preset |
| Custom images | Custom images and snapshots of your own sandboxes | Any public Docker Hub image for linux/amd64 |
| Self-hosting | No | Environments can run on your own infrastructure; on-premise on Enterprise |
| Compliance and transparency | SOC 2 is in progress. We're open about where we stand: live status and data retention policy. | Not published |
Every feature Battleships tracks (43)
| Feature | boat | OpenReward |
|---|---|---|
| Full VM (own kernel) | Yes | No |
| Docker inside | Yes | No |
| Nested virtualization | Yes | No |
| Root | Yes | Not published |
| Systemd | Yes | Not published |
| Desktop GUI | Yes | No |
| Browser | Yes | No |
| Computer-use API | Yes | No |
| SSH | Yes | No |
| Persistent disk | Yes | No |
| Pause/resume | Yes | No |
| Automatic snapshots | Yes | No |
| Fork/clone | Yes | No |
| Live fork (no pause) | Yes | No |
| 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 | Not published |
| Public IPv4 | No | No |
| Egress allowlist | No | No |
| Secret proxy | Yes | Yes |
| Secret proxy for any API | Yes | Yes |
| Preinstalled agents | Yes | No |
| Agent harness API | Yes | No |
| MCP server | No | No |
| GPU | No | Yes |
| Windows | No | No |
| Arm64 | No | No |
| Self-hosting | No | Yes |
| BYOC | No | Not published |
| Open source | No | No |
| Scoped API keys | Yes | Not published |
| Spend limits | Yes | Not published |
| Usage per sandbox (API) | Yes | No |
| Audit logs | No | Not published |
| SSO | No | Not published |
| SOC 2 | No | Not published |
| HIPAA | No | Not published |
| EU data residency | Yes | Not published |
| Zero data retention | Yes | Not published |
Performance
There is no comparable measurement for OpenReward Sandboxes. boat did 26.69 runs per second on the StarSling hpc-sandbox-benchmarks Node.js build-and-test loop at 4 vCPU / 8 GB, on dedicated AMD Ryzen 9 9950X hosts, and OpenReward has not been run on that benchmark.
Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. OpenReward 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 | OpenReward, 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 | $52 · 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 | $242 · 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 | $2,623 · 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 | $745 · Pay-as-you-go · with limits |
| 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 | $1,647 · Pay-as-you-go · with limits |
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.
OpenReward: OpenReward bills per second for the vCPU and RAM of the preset you pick, from prepaid organization credits, with the L4 GPU billed by the hour on top. A hosted environment also bills its own server for every session, so a sandbox used through an environment costs more than its own rate.
Limits in those scenarios
OpenReward does not fit the workday, always-on and agent fleet scenarios because a sandbox has no disk that survives a stop, and the fleet also needs Docker inside. For 300 or 1,000 machines at once, OpenReward publishes no per-account concurrency limit, so the page cannot confirm it fits. boat fits every scenario except the 1,000-job burst, where the $2,000 plan queues starts above 90 a minute. OpenReward also lists no disk size or egress price.
| Scenario | boat | OpenReward |
|---|---|---|
| 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 OpenReward when
you train or evaluate agents with RL on its hosted environments and rewards, need an L4 GPU in the task, or want to run environments on your own cluster.
- RL environments and rewards included. Teams training agents with RL get 380+ ready environments with verified rewards and a harness runner, behind one API.
- L4 GPUs and a no-network switch. Tasks that need a GPU can use an NVIDIA L4 preset, and one flag cuts a sandbox off from the internet.
- Run it on your own machines. Environments follow an open standard and can run on a laptop, your own cluster or OpenReward, with on-premise for Enterprise.
Use boat when
your agents need a real machine: Docker, emulators, a browser, a disk that stays between runs, forks of a running sandbox, and URLs or SSH to reach what they build.
- A whole VM, not a container. Agent tasks often run docker compose or build images. In an OpenReward container that is not documented, so those tasks fail, and nothing bigger than 4 vCPU is listed.
- A disk that stays, and forks. An OpenReward sandbox is deleted when its session ends, so an agent that stops loses its work, and every branch has to start over from the base image and redo its setup.
- Reach it from outside. OpenReward sandboxes only take run, upload and download calls, so an agent cannot show you a running app, a human cannot SSH in to help, and browser tasks have no screen.
Questions people ask
Is OpenReward a general sandbox like boat?
No. OpenReward's docs say its sandboxes are tied to an environment workspace and are not a general sandbox compute solution. They are containers that RL environments use to run agent code, and they are deleted when the session ends.
Is boat cheaper than OpenReward?
In the scenarios on this page boat costs less in four of five, but OpenReward does not fully fit any of them, mostly because its sandboxes keep no disk. For 1,000 short jobs at once neither fits cleanly, and OpenReward costs a little less there.
Can I run Docker inside an OpenReward sandbox?
OpenReward documents its sandbox as an isolated container and does not document Docker inside it. boat sandboxes are full VMs with Docker, systemd, root and /dev/kvm.
Can I use boat for RL rollouts?
Yes. Each rollout gets its own VM, you can fork a running sandbox to branch from the same state, and coding agents are preinstalled. boat does not host RL environments or rewards, so you bring your own tasks and grading.
Does either have GPUs?
OpenReward offers an NVIDIA L4 preset for sandboxes, and its docs say GPU capacity is limited. boat runs CPU machines only.
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