# CodeSandbox SDK vs Tensorlake: prices, limits and features.

**Both list about the same price for 2 vCPU and 4 GB: $0.15 / hour and $0.088 to $0.20 / hour.** Here is where they differ, from each one's own pages, plus a third option if neither fits.

## The short answer

The CodeSandbox SDK runs microVM sandboxes. Together AI now offers it as Together Code Sandbox and calls the old SDK legacy. Tensorlake runs sandboxes for agents as Firecracker microVMs with memory snapshots, and also runs GitHub Actions jobs.

### Pick CodeSandbox SDK if you want

- a lower price while the CPU is busy: $0.15 against $0.20 an hour

### Pick Tensorlake if you want

- a lower price while the CPU is idle: $0.088 against $0.15 an hour
- CPU billed only while busy; CodeSandbox SDK bills CPU the whole time a sandbox runs
- pause keeps memory: Yes, memory and running processes, which CodeSandbox SDK's pages do not state
- CI runners too: Yes, GitHub Actions runners (CodeSandbox SDK: No)
- fork and snapshots: Memory snapshots and warm copies, which CodeSandbox SDK's pages do not state
- self-host: In your VPC or on-prem, on Enterprise, which CodeSandbox SDK's pages do not state
- GPUs: H100, by request, which CodeSandbox SDK's pages do not state
- security report: SOC 2 Type 2; HIPAA on Pro, which CodeSandbox SDK's pages do not state

### Or Groundbox if you want

- a lower list price than both: $0.08 / hour

Where CodeSandbox SDK or Tensorlake is ahead of Groundbox: longest session, self-host, GPUs, security report.

## Side by side

| Feature | CodeSandbox SDK | Tensorlake | Groundbox |
| --- | --- | --- | --- |
| Price, 2 vCPU · 4 GB | [$0.15 / hour](https://www.together.ai/pricing) | [$0.088 to $0.20 / hour](https://docs.tensorlake.ai/platform/billing.md) | $0.08 / hour |
| Pause keeps memory | Not published | [Yes, memory and running processes](https://docs.tensorlake.ai/sandboxes/lifecycle.md) | Yes, memory and disk |
| CI runners too | [No: the pricing page lists Code Sandbox and Code Interpreter only](https://www.together.ai/pricing) | [Yes, GitHub Actions runners](https://docs.tensorlake.ai/github-actions/quickstart.md) | Yes, on the same key and bill |
| Isolation | [microVM](https://docs.together.ai/docs/together-code-sandbox-legacy.md) | [Firecracker microVM](https://www.tensorlake.ai/pricing) | Firecracker VM |
| Fork and snapshots | Not published | [Memory snapshots and warm copies](https://docs.tensorlake.ai/sandboxes/snapshots.md) | Fork a running sandbox; save points |
| Longest session | [Runs until stopped, unless a ttl is set](https://docs.together.ai/docs/together-code-sandbox.md) | [No limit on Pro; 24 hours on credits, 2 hours free](https://docs.tensorlake.ai/sandboxes/lifecycle.md) | 720 hours, renewable |
| Self-host | Not published | [In your VPC or on-prem, on Enterprise](https://www.tensorlake.ai/pricing) | No |
| GPUs | Not published | [H100, by request](https://docs.tensorlake.ai/sandboxes/gpu.md) | Not offered |
| Largest size | [16 cores](https://docs.together.ai/docs/together-code-sandbox.md) | [16 vCPU on Pro](https://docs.tensorlake.ai/sandboxes/lifecycle.md) | 16 vCPU |
| SDK languages | [Python and TypeScript](https://docs.together.ai/docs/together-code-sandbox.md) | [Python and TypeScript](https://docs.tensorlake.ai/sandboxes/quickstart.md) | TypeScript and Python |
| Security report | Not published | [SOC 2 Type 2; HIPAA on Pro](https://www.tensorlake.ai/pricing) | No SOC 2 report |
| Free to start | [Legacy SDK: a free Build plan](https://docs.together.ai/docs/together-code-sandbox-legacy.md) | [1 sandbox, 2-hour sessions, no card](https://www.tensorlake.ai/pricing) | $5 of usage, no card |
| Plans | [Legacy SDK: Build (free) and Scale plans; the new SDK is sold through Together's sales team](https://docs.together.ai/docs/together-code-sandbox-legacy.md) | [Free; prepaid usage credits from $5; Pro $250 / month](https://www.tensorlake.ai/pricing) | Free: everything, with $5 of usage. $20 / month: $20 of usage included, then pay as you go |
| CPU | [$0.0446 per vCPU-hour (Together)](https://www.together.ai/pricing) | [$0.07 per active core-hour on credits, $0.042 on Pro; at least 20% of the cores is billed](https://docs.tensorlake.ai/platform/billing.md) | $0.04 per vCPU-hour, a dedicated core |
| Memory | [$0.0149 per GiB-hour (Together)](https://www.together.ai/pricing) | [$0.015 per GB-hour on credits, $0.009 on Pro, billed apart from CPU](https://docs.tensorlake.ai/platform/billing.md) | Included, up to 4 GB per vCPU |
| Billing | [Legacy SDK: VM credits at $0.01486 each, rounded up to the minute](https://docs.together.ai/docs/together-code-sandbox-legacy.md) | [Per second](https://docs.tensorlake.ai/platform/billing.md) | Per second, no minimum |
| Sizes | [New SDK: 0.1 to 16 cores, 1 to 8 GB per CPU](https://docs.together.ai/docs/together-code-sandbox.md) | [Up to 4 vCPU on credits and 16 on Pro, 1 to 8 GB per vCPU](https://docs.tensorlake.ai/sandboxes/lifecycle.md) | 1 to 16 vCPU, up to 4 GB per vCPU |
| Sandboxes at once | [Legacy SDK: 10 on Build, 250 on Scale](https://docs.together.ai/docs/together-code-sandbox-legacy.md) | [1 free, 100 on credits, 1,000 on Pro](https://www.tensorlake.ai/pricing) | 20 once a card is on file |
| Egress controls | Not published | [Allow and deny lists by domain, IP or CIDR, changeable while running](https://docs.tensorlake.ai/sandboxes/networking.md) | An allowlist of hosts per sandbox |

a rust mark means that side is ahead · each value links to its source · checked 2026-11-20

Each against Groundbox: [Groundbox vs CodeSandbox SDK](https://groundbox.dev/compare/codesandbox) and [Groundbox vs Tensorlake](https://groundbox.dev/compare/tensorlake). Every other comparison is on [/compare](https://groundbox.dev/compare).

## Sources

### CodeSandbox SDK

- [Together: Code Sandbox (legacy CodeSandbox SDK)](https://docs.together.ai/docs/together-code-sandbox-legacy.md), checked 2026-11-19
- [Together: Code Sandbox](https://docs.together.ai/docs/together-code-sandbox.md), checked 2026-11-19
- [Together pricing](https://www.together.ai/pricing), checked 2026-11-20

### Tensorlake

- [Tensorlake pricing](https://www.tensorlake.ai/pricing), checked 2026-11-20
- [Tensorlake docs](https://docs.tensorlake.ai/), checked 2026-11-20
- [Tensorlake docs: billing](https://docs.tensorlake.ai/platform/billing.md), checked 2026-11-20
- [Tensorlake docs: sandbox lifecycle](https://docs.tensorlake.ai/sandboxes/lifecycle.md), checked 2026-11-20
- [Tensorlake docs: snapshots](https://docs.tensorlake.ai/sandboxes/snapshots.md), checked 2026-11-20
- [Tensorlake docs: networking](https://docs.tensorlake.ai/sandboxes/networking.md), checked 2026-11-20
- [Tensorlake docs: GPU sandboxes](https://docs.tensorlake.ai/sandboxes/gpu.md), checked 2026-11-20
- [Tensorlake docs: quickstart](https://docs.tensorlake.ai/sandboxes/quickstart.md), checked 2026-11-20
- [Tensorlake docs: GitHub Actions](https://docs.tensorlake.ai/github-actions/quickstart.md), checked 2026-11-20

## Try the third option free.

[Start free, no card](https://groundbox.dev/signup)[Groundbox vs CodeSandbox SDK](https://groundbox.dev/compare/codesandbox)
