The AI builder’s cloud
Everything you need to train, fine-tune, and serve your models on Crusoe Cloud.
Connect with us
at events
Your toolkit
Demos
Real examples from our GitHub and Hugging Face Spaces. Run the demos, fork the code.
Model Arena
Pit open-source LLMs against each other, same prompt, blind results, real latency. Vote on winners, compare costs vs. proprietary APIs, and simulate failover across a 3-step agent pipeline. All running on Crusoe Intelligence Foundry.
FakePod
Throw two LLMs into a podcast. Qwen3-235B and Nemotron-120B take turns riffing on any topic you give them, with distinct personalities, neural text-to-speech voices, and live subtitles. Conversations run up to 20 minutes. Hit stop whenever.
Snake Battle AI
Pick two open models. Drop them into a snake game. Watch them think, move, and crash in real time. DeepSeek vs. Llama? Qwen vs. Gemma? The board decides who's actually smarter.
TorchTitan Pretraining on CMK
This example runs a two-node, 16-GPU pre-training job for Meta's Llama 3.1 8B model using TorchTitan on Crusoe Managed Kubernetes, orchestrated via the Kubeflow Training Operator PyTorchJob CRD.
PyTorch Distributed Training on SLURM
Fork PyTorch distributed training examples for multi-node, multi-GPU jobs on Crusoe Cloud Slurm clusters: Fully Sharded Data Parallel (FSDP), Distributed Data Parallel (DDP), torchtitan, and vision transformers.
Ray Inference on CMK
Clone example deployments for serving LLMs on Crusoe Managed Kubernetes with vLLM, Ray Serve, and the KubeRay operator for single-node and multi-node inference.
From our engineering team
Frequently
asked questions
Crusoe Intelligence Foundry supports a growing catalog of leading open models, such as NVIDIA Nemotron, Google Gemma, DeepSeek, and Meta Llama, all available through Crusoe Managed Inference. Need a model that's not listed? Bring your own fine-tuned model to optimize performance for your custom checkpoint. The catalog is continuously expanding, so check back regularly or reach out for early access to upcoming models.
Yes. Crusoe offers fully managed Slurm running natively on Crusoe Managed Kubernetes (CMK), built on SchedMD's Slinky v1.0. Researchers get Slurm commands they already know (sbatch, srun, squeue, sinfo) via dedicated SSH login nodes. Kubernetes administrators manage the cluster lifecycle through kubectl or the Crusoe console, so researchers don’t need to touch Kubernetes directly. Key capabilities include topology-aware scheduling using NVIDIA topograph, automated job re-queueing, GPU-ready container images pre-loaded with CUDA, cuDNN, NCCL, and the full InfiniBand stack. It also integrates with Crusoe AutoClusters for automated GPU health monitoring and node replacement. Because Slurm runs on the same foundation as CMK, every platform capability works out of the box, from Command Center observability to automated node remediation. Read the full engineering deep-dive to learn how it was built.
Pricing is flexible, transparent, and built for AI workloads. GPU and CPU instances are available on-demand (billed per minute), spot (up to 90% off for fault-tolerant workloads), or reserved (the deepest discounts). Crusoe Managed Inference is pay-as-you-go per million tokens, or provisioned throughput via AI Model Units (AMUs). No setup fees, no hidden costs, and no data transfer charges. Contact sales to design a commitment that fits your roadmap.
Yes. Crusoe Managed Kubernetes (CMK) supports standard Kubernetes APIs, kubectl, Helm, and programmatic access. It’s fully compatible with GitHub Actions, GitLab CI, Jenkins, ArgoCD, and your existing MLOps pipelines. Command Center streams infrastructure metrics into your observability stack via Telemetry Conduit, with out-of-the-box logging for Kubernetes and journald. Explore the Crusoe Cookbook for hands-on integration guides.
Crusoe Cloud holds SOC 2 Type II, ISO 27001, and ISO 42001 certifications. Customer data is protected by technical and organizational safeguards against unauthorized access, loss, and alteration. All certifications, audit reports, and security documentation are available through the Crusoe Trust Center. Data processing terms are published at the Legal Center.
Get 24/7 enterprise-grade support with under 6-minute average first reply and 100% CSAT for 12+ months. Developer resources include full documentation, support portal, the engineering blog, and the Crusoe Cookbook.
As a developer, you can access Crusoe Cloud via Crusoe Cloud Console, CLI, REST API, Terraform or Crusoe Cloud MCP Server.



.png)



