Crusoe has announced Serverless Fine-Tuning and Self-Serve Deployments in Crusoe Intelligence Foundry, the managed AI platform for Crusoe Cloud. These capabilities give data scientists and ML engineers a complete path from proprietary data to production-ready models, on purpose-built AI infrastructure, without the overhead of managing it.
Fine-tuning is now a standard part of building with open-source AI models, and as open-weight models catch up to proprietary models, more teams are choosing to customize with their proprietary data and retain ownership of the fine-tuned weights. Getting started is straightforward, but doing it repeatedly adds up. Idle clusters, hardware failures, and scattered tools slow teams down, and the people who should be improving the model end up troubleshooting infrastructure instead.
Crusoe Serverless Fine-Tuning: From data to deployed model in minutes
Crusoe Serverless Fine-Tuning eliminates infrastructure overhead so engineering teams can focus on model quality. Teams can launch a fine-tuning job in a few clicks — select a base model from a curated library of top-performing open-weight models, upload a custom data set, configure settings with pre-configured best practices, and submit; no dedicated reservation required.
Jobs run on Crusoe’s distributed AI-optimized infrastructure with automated recovery and restart if hardware blips are detected. As soon as the model stops improving, teams stop paying. When tuning is complete, model weights are returned in portable .safetensors format. They can go live in one click using Crusoe’s new Self-Serve Deployments for inference, or the weights can be downloaded to deploy anywhere.
“Our early experience with Crusoe’s Serverless Fine Tuning product was seamless, and it worked like a charm. We look forward to leveraging it to optimize the latency and cost of our AI agents as we scale our infrastructure,” said Dr. Will Leeney, and Dr. Hiskias Dingeto, AI Researchers, StackOne.
“Open models have definitely crossed the quality threshold, while offering unique optimization opportunities with your data, and giving you full control of their lifecycle, ” said Erwan Menard, Senior Vice President of Product, Crusoe Cloud.
“With Crusoe Serverless Fine-Tuning and Self-Serve Deployments your journey just got easier; fast iteration, predictable cost, and the guarantee that your data and weights stay yours. You shouldn’t have to choose between a managed experience and ownership of your model.”
Crusoe Self-Serve Deployments: From fine-tuning to flexible inference in one click
Crusoe Cloud is expanding the consumption options available for Crusoe Managed Inference and adding Self-Serve Deployments. While Serverless Inference APIs are ideal for early-stage experimentation, Self-Serve Deployments is designed for teams with production-ready workloads.
Billed by GPU per hour for predictable costs, users can select a base model in Intelligence Foundry, choose an inference profile optimized for throughput or responsiveness, and deploy to production on NVIDIA H100 or H200 GPUs—without ever touching the underlying infrastructure.
For teams running continuous post-training loops with Serverless Fine-Tuning in Intelligence Foundry, the ability to tune models and deploy to a production inference endpoint in one click eliminates the friction of moving model artifacts between providers.
Self-Serve Deployments expands the inference options available in Crusoe Intelligence Foundry. Customers can now choose Serverless Inference APIs for quick experimentation, Self-Serve Deployments for production-ready workloads optimized for throughput or responsiveness, or Tailored Deployments for dedicated, custom inference on any fine-tuned or proprietary model with SLA-backed performance.
Teams building the next generation of AI products, like Yutori, Nous Research, Wonderful, Salient, Composite, and Magicare, run on Crusoe because fast, reliable inference optimized to their stack is how they deliver performance to the people who depend on them.
Key capabilities at general availability
Serverless Fine-Tuning:
- Developer-friendly UI, SDK, and API
- Curated library of top performing base model families, including Qwen, DeepSeek, Gemma, gpt-oss, and more
- LoRA (low-rank adaptation) fine-tuning for lightweight customization that delivers fast, cost-efficient iteration
- Automatic job recovery and restart, plus checkpoints that save at every step – early stopping ends billing the moment the model stops improving
- Full job lineage: every tuned artifact traces back to the exact data and configuration that produced it
- Native export of raw weights in .safetensors format
Self-Serve Deployments:
- Predictable performance for production-grade workloads
- Inference profiles optimized for throughput or responsiveness based on your unique needs
- OpenAI-compatible API for zero-friction integration with existing applications
- One-click deployment from Serverless Fine-Tuning workflow
