Available models - CoreWeave Docs

Serverless Training supports the following foundation models for both reinforcement learning (RL) and supervised fine-tuning (SFT). To express interest in a particular model that isn’t in the model catalog, contact support.

Model catalog

Model	Model ID (for API usage)	Type	Context Window	Parameters	Description
Google Gemma 4 26B A4B Instruct	`google/gemma-4-26B-A4B-it`	Text, Vision	262k	4B-26B (Active-Total)	Gemma 4 26B A4B is a multimodal MoE model with LoRA support and function calling for agentic workflows.
Meta Llama 3.1 8B	`meta-llama/Llama-3.1-8B-Instruct`	Text	131k	8B (Total)	Efficient conversational model optimized for responsive multilingual chatbot interactions.
NVIDIA Nemotron 3.5 Lightning	`nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B`	Text	262k	3B-30B (Active-Total)	Nemotron 3.5 Lightning is an MoE model built for fast, reliable agentic tasks across use cases such as financial services, cybersecurity, telecom, and retail.
Qwen3.8 27B	`Qwen/Qwen3.8-27B`	Text, Vision	262k	27B (Total)	Qwen3.8-27B is a dense multimodal model suited for coding, research, vision, and long-running agent tasks.
Qwen3.6 35B A3B	`Qwen/Qwen3.6-35B-A3B`	Text, Vision	262k	3B-35B (Active-Total)	Qwen3.6-35B-A3B is an MoE multimodal model with 262K context optimized for agentic coding workflows.

LoRA rank and alpha

Serverless Training sets the LoRA rank and alpha for each base model. You can’t change them using the API or ART. The rank depends on the base model, not on the training method.

Model ID (for API usage)	LoRA rank
`google/gemma-4-26B-A4B-it`	1
`meta-llama/Llama-3.1-8B-Instruct`	8
`nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B`	8
`Qwen/Qwen3.6-35B-A3B`	1
`Qwen/Qwen3.8-27B`	8

To check the rank and alpha of an adapter you trained, open the adapter_config.json file in its lora artifact and read the r and lora_alpha values.

​Model catalog

​LoRA rank and alpha

Model catalog

LoRA rank and alpha