Supported models

SambaNova Cloud API provides access to DeepSeek models, the Llama 3.3, 3.2, and 3.1 family of models, as well as the Qwen 2.5 family, all at full precision. SambaNova provides access to all models for all tiers.

Model details

A structured comparison of various models across the Qwen and Llama families is provided below. It describes key specifications such as model name, ID, developer, context length, and provides links to their respective model cards.

Preview models

Preview models in SambaNova Cloud are available as early-access offerings intended primarily for evaluation purposes. During the preview phase, these models have limited capacity but are fully functional in terms of accuracy and performance.

Developer	Model ID	Context length	View on Hugging Face
DeepSeek
	DeepSeek-V3-0324	8k tokens	Model card
OpenAI
	Whisper-Large-V3	N/A	Model card
Meta
	Llama-4-Scout-17B-16E-Instruct	8k tokens	Model card
	Llama-4-Maverick-17B-128E-Instruct	8k tokens	Model card
Qwen
	Qwen2-Audio-7B-Instruct	N/A	Model card

Production models

Production models meet our high standards for speed and quality and are intended for use in production environments.

Developer	Model ID	Context length	View on Hugging Face
DeepSeek
	DeepSeek-R1	16k tokens	Model card
	DeepSeek-R1-Distill-Llama-70B	128k tokens	Model card
Meta
	Meta-Llama-3.3-70B-Instruct	128k tokens	Model card
	Meta-Llama-3.2-3B-Instruct	8k tokens	Model card
	Meta-Llama-3.2-1B-Instruct	16k tokens	Model card
	Meta-Llama-3.1-405B-Instruct	16k tokens	Model card
	Meta-Llama-3.1-8B-Instruct	16k tokens	Model card
	Meta-Llama-Guard-3-8B	8k tokens	Model card
Qwen
	QwQ-32B	16k tokens	Model card
Tokyotech-llm
	Llama-3.1-Swallow-8B-Instruct-v0.3	16k tokens	Model card
Other
	E5-Mistral-7B-Instruct	4k tokens	Model card