XOps 16
For short-term testing, small quantized models, and lightweight builds.
- Per day$19.4
- Per week$52.3
- Per month$96.8
- Per quarter$263.3
All three plans run on dedicated physical machines. CPU, GPU, and unified memory are never shared with other tenants. Prices are in USD, prepaid by the day, week, month, or quarter. Choose the memory you need, then pick a term that fits how long you’ll run it.
Start small-scale quantized model experiments with 16GB. For models you want to keep serving, consider 24GB. For larger models or parallel workloads, compare 64GB. For builds, choose based on project size and cache needs—there’s no need to pay for memory you won’t use.
Starting at $19.4/day · Usually ready in about 4 minutes · Availability is confirmed in real time in the console
Choose by unified memory
Model weights, context, and runtime all use memory. Use the capacities below as a starting point, not a compatibility guarantee. Check your model files and actual peak usage before deployment.
Switching terms changes only the headline price on each card; all four prices are listed below it. Add-ons are charged separately. Check the console for your checkout total.
For short-term testing, small quantized models, and lightweight builds.
Run popular 7B–8B quantized models continuously, with room for the service to operate.
For larger models, parallel inference, or teams that need more local build cache.
Every plan comes on a dedicated physical machine with its own public IP. Check the checkout page for actual bandwidth and available network options.
| Plan | Chip | Unified memory | Storage | Per day |
|---|---|---|---|---|
| XOps 16 | M4 | 16GB | 256GB | $19.4 |
| XOps 24 | M4 | 24GB | 512GB | $40.8 |
| XOps Pro 64 | M4 Pro | 64GB | 2TB | $61.2 |
Use these recommendations as a starting point. Quantization, context length, project dependencies, and concurrent workloads all affect peak usage. Test with your own workload before putting a plan into production.
16GB suits small quantized model experiments. 24GB is a good starting point for serving popular 7B–8B quantized models. Choose 64GB for larger models or parallel inference. Budget memory for weights, context cache, and other processes.
Run a real request first, then check peak memory usage.
Start by testing XOps 16 for small projects and a single runner. Consider XOps 24 for projects with more dependencies or when you need to retain build caches. For parallel builds and tests, or switching between multiple projects, consider XOps Pro 64. The machine’s resources are not shared with other tenants.
Track the time for one clean build and one incremental build.
Check your project files, sample libraries, and export space first. Try the entry plan for light editing. For large sample libraries, consider a storage add-on; for complex offline exports, watch CPU and memory peaks. If you’ll use remote desktop, factor in network performance too.
Storage add-ons are priced separately from the base plan.
Prices below are in USD per add-on and rental term; they do not include the base plan. Thunderbolt 5 linking is priced per machine. Check the checkout page for available options.
| Add-on | Per day | Per week | Per month | Per quarter |
|---|---|---|---|---|
| +1TB SSD | $2.3 | $6.1 | $11.3 | $30.7 |
| +2TB SSD | $4.6 | $12.2 | $22.6 | $61.4 |
| Thunderbolt 5 link (per machine) | $1.2 | $3.2 | $5.9 | $16 |
Before your model weights, sample libraries, or build cache fill the base storage, estimate your working set and temporary-file peaks. Extra storage does not replace unified memory.
All three plans are offered in Singapore, Tokyo, Seoul, and Hong Kong. The table shows catalog availability; check the console for real-time availability when you order.
| Plan | Singapore | Tokyo | Seoul | Hong Kong |
|---|---|---|---|---|
| XOps 16 | Available | Available | Available | Available |
| XOps 24 | Available | Available | Available | Available |
| XOps Pro 64 | Available | Available | Available | Available |
“Available” means the plan can be ordered from the catalog; it is not a real-time inventory snapshot. Cross-region latency depends on your network provider and the time of day. Before choosing a location, see the tests in our connection guide.
All charges are in USD. Pay with USDT-TRC20 or a Visa, Mastercard, or Amex card processed by Stripe. Available payment gateways at checkout are confirmed by the backend.
Choose a plan, term, location, and add-ons on the configuration page, then review your order total.
Sign in to the console to view your order and billing, and renew or change add-ons where available.
Release the machine from the console when you no longer need it. Move your projects, model weights, and caches first, then check your order status.
Order actions depend on the options available in the console and the terms for your order. To check a specific order, sign in and submit a support ticket.
Each plan uses different physical hardware, so memory can’t be increased like a software quota. Before switching, check your current order and the new plan’s availability in the console, plan your data migration, and submit a support ticket to confirm how your order can be handled.
Check your order’s end date and renewal options in the console, and renew or move your data before it expires. Unrenewed instances are released at the end of the term in accordance with the service terms. Don’t rely on the instance disk as your only backup.
Move your data off the add-on storage first, then check the available order changes in the console. Add-on changes and when they take effect are shown on the order page. Storage in use cannot be assumed safe to remove.
The console provides order and payment records so you can check charges by term, plan, and add-on. To ask about a specific billing field or payment transaction, sign in and submit a support ticket with your order number.
Choose a plan and location, then pay. Your machine is usually ready in about 4 minutes. Test it with your actual workload before committing to a longer term.