Top open models, production-ready
Browse open models from leading labs available for managed inference and fine-tuning. No infrastructure to manage.
Models
Beam 1.0 501B A23B
DeepSeek V4.1 Flash 1017
DeepSeek V4 Flash 0731
Gemma 4 26B A4B
GLM-5.2
GPT-OSS 20B
Inkling Small
Inkling Small FP8 Dynamic
Llama 3.3 70B Instruct
Qwen3 235B A22B Instruct
Qwen3.5 2B
Qwen3.5 35B A3B
Qwen3.5 9B
Qwen3.5 9B Base
Qwen3.6 27B
Qwen3.6 35B A3B
Qwen3.6 35B A3B FFT
Qwen3.8 27B FFT
Qwen3 8B
Nemotron 3.5 Super VL 120B A12B
GLM-5.3 Flash
GLM-5.3
DeepSeek V4.1 Flash
Gemma 4 31B
GPT-OSS 120B
Llama 3.1 8B Instruct
Nemotron 3.5 Lightning 30B A3B
Nemotron 3 Nano 30B
Nemotron 3 Nano Omni
Nemotron 3 Super 120B
Nemotron 3 VoiceChat
Qwen3.5 4B
Qwen3.8 27B
Fine-tune and deploy in Crusoe Intelligence Foundry
Explore model providers
Frequently
asked questions
Choosing a model starts with the task, then narrows by four properties: model type, parameter count, context length, and license. Crusoe's model library lets you filter by type, capability, and provider, and every tile shows parameters, context window, per-token price, and what services you can use each model with: Serverless Inference, Self-Serve Deployments, and/or Serverless Fine-Tuning. Shortlist two or three candidates, test them using the playground in Crusoe Intelligence Foundry with real prompts, then commit dedicated capacity to the winner.
The models in Crusoe's library are open models, meaning their weights are published and licensed for use, though most are open-weight rather than open-source, in the strict sense of also releasing training code and data. Licenses range from permissive MIT and Apache 2.0 terms to community licenses that carry an acceptable use policy. Crusoe lists the license and any acceptable use policy on every model page so you can confirm commercial terms before deploying.
Every model in Crusoe's library shows the price per million tokens on the tile, and how you are billed depends on the way you run it. Serverless Inference and Serverless Fine-Tuning bill per token, so cost tracks usage with no reserved capacity. Self-Serve Deployments bill per GPU-hour for dedicated capacity, which brings cost per token down at sustained volume. Current rates are also listed on the Crusoe Cloud pricing page.
Crusoe's model library spans the main categories of open models used in production: instruct models for chat and agents, reasoning models for multi-step tasks, multimodal models that take images or audio alongside text, and speech models for voice applications, across small and large parameter sizes. Every model is tagged by type and capability so you can filter the library to what fits your workload, and each tile shows which managed AI services support it.
An open model publishes its weights so anyone can download, run, fine-tune, and deploy it on their own infrastructure under the terms of its license, while a closed model is accessible only through its provider's API and cannot be inspected, modified, or moved. Most open models are open-weight rather than fully open-source, since training code and data usually stay private. Every model in Crusoe's library is an open model, with its license and any acceptable use policy shown on the model page.



%204.png)