Crusoe
Crusoe Cloud publishes on-demand GPU hourly rates and per-1M-token charges for managed inference and serverless fine-tuning, plus self-serve deployment endpoints and a small free Foundry credit. Inputs are model artifacts and data; actions are fine-tuning and serving; outputs are hosted model endpoints billed by GPU-hours or tokens.
Crusoe lists pay-as-you-go GPU hourly rates and per-1M-token rates for Serverless Fine-Tuning and Serverless Inference; some GPU models and enterprise volume/commitment rates require contacting sales.Is Crusoe right for you?
Good for
- Founders who need transparent per-hour GPU pricing and per-token managed inference or fine-tuning charges to model compute costs for experiments and prototyping.
- Public per-GPU hourly pricing for several GPU models enables precise short-term cost estimates.
- Per-1M-token pricing is published for Serverless Fine-Tuning and Serverless Inference for many models.
Keep in mind
- Many GPU models and enterprise volume/reserved rates are not shown as fixed prices and require sales engagement.
- Some GPU rows display 'Contact sales' instead of a published hourly rate.
- Buyers who expect full long-term reserved or discounted enterprise commitment rates published on the site without contacting sales, since those rates require engagement.
Pricing and plan limits
The site shows specific hourly and per-1M-token prices for many items (for example NVIDIA H200 $4.29/GPU-hr and several token rates). For reserved, volume, or tailored deployments the pages instruct you to contact sales rather than publish fixed prices.
More about capabilities and limits
Documented strengths
Public per-GPU hourly pricing for several GPU models enables precise short-term cost estimates.
Per-1M-token pricing is published for Serverless Fine-Tuning and Serverless Inference for many models.
The Foundry offers a small free credit to try fine-tuning and serving models.
Limitations to consider
Many GPU models and enterprise volume/reserved rates are not shown as fixed prices and require sales engagement.
Some GPU rows display 'Contact sales' instead of a published hourly rate.
Choosing a plan
Run a short proof-of-concept on the listed GPU or serverless model to measure real hourly and token costs, then request volume/commitment pricing from sales for longer-term needs.
Based on the linked product documentation.
Available on
Web.
Official sources
Alternatives to Crusoe
Choose around the work you need to do.
DeepInfra
Use DeepInfra when you need pay-as-you-go per-inference (per-1M-token) billing alongside the option to rent dedicated GPUs by the hour for predictable inference capacity.
Baseten
Choose Baseten for deploying and scaling open-source or fine-tuned models with managed or self-hosted options and documented per-token and per-minute compute pricing for production inference.
Together AI
Consider Together AI when you want a platform offering serverless inference plus provisioned and dedicated GPU capacity with per-model token and per-GPU pricing for production deployments.