GPU clouds / SF Tensor Model Foundry alternative

An SF Tensor alternative for the agent around model training

SF Tensor currently offers Model Foundry and embedded research engagements for training models across different accelerator platforms. Its Tensor Cloud launch described job placement and result delivery across providers. These are training and optimization services, with different boundaries from a persistent general-purpose agent computer.

Start buildingCompare capabilities ↓

Reviewed 2026-10-19 · Published by the Sandbox0 team

When SF Tensor Model Foundry fits

SF Tensor fits teams needing training infrastructure, architecture-specific kernel work, cross-accelerator execution or embedded model researchers. Keep required training services, model expertise and accelerator capacity.

When to consider Sandbox0

Consider Sandbox0 for compatible Linux agents that prepare data, build tools or orchestrate training and need the complete durable writable RootFS, automatic online growth and supported files while guest compute stays paused.

Capabilities

SF Tensor Model Foundry and Sandbox0: what persists?

Feature comparison with documentation and verification status
Environment and runtime lifetimeSF Tensor Model FoundryNot established in this reviewSandbox0ttl and hard_ttl default to 0 (disabled). Configure idle pause and deletion policies for the same persistent sandbox identity; an idle running sandbox still accrues compute charges.Read documentation →
Persistent filesSF Tensor Model FoundryResults can be delivered to buckets or provider persistent storage; a universal retained writable guest disk is not established.View evidence ↗Sandbox0The writable RootFS survives successful checkpoints, including software, system configuration, dependencies, and user data.Read documentation →
Whole RootFSSF Tensor Model FoundryCross-provider job execution and retained results do not establish complete mutable OS filesystem continuity.View evidence ↗Sandbox0Persistence extends beyond /workspace to paths such as /usr/local, /etc, and /home. Runtime tmpfs and virtual filesystems are excluded.Read documentation →
Release compute, keep stateSF Tensor Model FoundryTraining recovery and job placement do not establish a user-controlled durable agent-machine pause/resume contract.View evidence ↗Sandbox0Pause compute between tasks; resume on available capacity. Default resume starts fresh processes from the saved filesystem.Read documentation →
Memory and process stateSF Tensor Model FoundryThe current reviewed material does not establish arbitrary guest process/RAM snapshot restoration.View evidence ↗Sandbox0Experimental memory checkpoints can preserve compatible process state. Use memory: true on both pause and resume on a compatible deployment. A filesystem checkpoint alone does not preserve running processes.Read documentation →

Online RootFS growth

On growth-enabled images, Sandbox0 automatically expands the writable RootFS as free space runs low, within a configured sparse logical ceiling up to 64 TiB. Agent processes keep running. Growth is enabled in the qualified Cloud region; confirm activation for other regions. Existing images and snapshots retain their captured storage geometry. The ceiling is logical capacity, not preallocated host disk.

SF Tensor’s training scale and provider selection concern accelerator workloads. They do not establish automatic online expansion of a complete durable per-agent RootFS.

Files while compute is paused

Sandbox0's Files API reads persistent files without resuming compute. Filesystem-only pauses also permit edits; retained-memory pauses are read-only and reject mutations with 409. SSH, SFTP, file watching, and runtime-only mounts require running compute.

See paused file operations →

The current offer is scoped around training outcomes

Model Foundry uses savings-share pricing. A separate embedded-research engagement has a fixed scope with compute billed separately. Neither arrangement is a public fixed hourly tariff for an arbitrary agent VM, and advertised performance gains are not a guaranteed saving for every workload.

Source 1 ↗

Cloud result retention is a selected storage path

The Tensor Cloud launch describes delivering results to a storage bucket, downloading them, or using the selected provider’s persistent storage. Those output destinations do not establish retention of every modified operating-system path or process memory across job moves. The current terms place data protection and backup responsibility with the customer.

Source 1 ↗Source 2 ↗

Confirm the savings baseline and eligibility in the agreement

Published Tensor Cloud terms list savings-fee percentages of 10%, 8% and 6% across rolling annual-spend bands, with underlying actual costs separate. Their written definition of Savings is inconsistent with the homepage’s lower-bill positioning, so confirm the calculation and reference baseline in the effective agreement. The published terms also restrict service eligibility to US residents and US-organized entities.

Source 1 ↗Source 2 ↗

Pricing

Compare billing models, then price your workload

CPU allocation, memory, billing minimums, disk, plan fees, and supported workload differ. These meters are not equal-performance machine quotes. Confirm region, resource units, and included allowances before estimating a total.

SF Tensor Model Foundry

Pricing reviewed 2026-10-19

Model Foundry
Scoped savings-share agreement
Confirm the savings definition, baseline and applicable percentage.
Rate source ↗
Published Tensor Cloud fee bands
10% / 8% / 6% of defined Savings
Rolling annual spend: up to $50,000; above $50,000 through $200,000; above $200,000. Custom written terms can supersede.
Rate source ↗
Underlying cloud resources
Actual compute and infrastructure costs, separately payable
Rate source ↗
Embedded research engagement
Fixed engagement quote; compute billed separately
Rate source ↗

No fixed public GPU-hour price is implied by a savings-fee percentage. Confirm the current product, US eligibility, reference prices, actual resource costs and the effective savings formula before estimating an invoice.

Check current SF Tensor Model Foundry pricing ↗

Sandbox0

$0.015/GiB-hour compute (billed by the second at approximately $0.000004167/GiB-second), plus $0.02/GiB-month of retained incremental RootFS and memory checkpoint data. Shared base images and platform recovery retention are excluded; paused sandboxes accrue storage charges only.

Compute / GiB-hour
$0.015
Retained data / GiB-month
$0.02
Paused compute
$0.00

A 4 GiB agent, 100 running hours

$6.00 compute

4 GiB × 100 hours, billed by the second. Keeping 20 GiB of billable incremental data for a full billing month adds $0.40. CPU is derived from memory by the platform; choose a tier that fits your workload.

Understand storage billing →

Before migrating from SF Tensor Model Foundry

  1. Inventory training jobs, accelerator dependencies, output destinations, cloud profiles and the effective savings or engagement agreement.
  2. Keep needed training services and export required model/data artifacts and compatible Linux toolchains.
  3. Test agent system retention, growth, supported paused files and orchestration separately from training performance and savings calculations.

Common questions

Is Sandbox0 a drop-in SF Tensor Model Foundry replacement?

Evaluate API compatibility and product scope. These products serve different core workflows, so a change may require a different architecture.

Does keeping files also keep a running agent?

Filesystem state and volatile process state are separate. Sandbox0 requires explicit compatible memory checkpoints to preserve running processes; its default filesystem resume starts fresh processes.

Sources and review scope

The Boat team's Battleships directory is the discovery source. References below describe the product and plan discussed on this page. Check current availability, regional terms and billing units before migrating. Where the cited material does not establish a capability or price, the comparison identifies that uncertainty.

Battleships research record ↗sf-tensor.com/ ↗sf-tensor.com/news/tensor-cloud-launch ↗sf-tensor.com/legal/terms ↗Sandbox0 RootFS capacity →

Related comparisons

Azure GPU VMs (NC_A100_v4 reference) vs Sandbox0 →Cerebrium Serverless Compute vs Sandbox0 →Crusoe vs Sandbox0 →DigitalOcean GPU Droplets vs Sandbox0 →Fluidstack vs Sandbox0 →Google Compute Engine GPU VMs vs Sandbox0 →