Model Catalog
Copy the complete
docker runfrom the Intel® Software Catalog. Open the model you want and use the Deployment tab.
This table is informational — use it to look up the Docker image name.
| Model | Description | License | HF token required | Tags | Docker Image |
|---|---|---|---|---|---|
| google/gemma-4-E4B-it | Multimodal, instruction-tuned model from Google's Gemma family with the E4B (Effective 4B) architecture, supporting text and image inputs for chat. | Apache-2.0 | No | text-generation, chat, instruction, multimodal | intel/inference-xeon-google-gemma-4-e4b-it:0.1.0 |
| ibm-granite/granite-3.2-2b-instruct | 2B-parameter instruction-tuned model from IBM's Granite 3.2 series, with a built-in toggle for extended chain-of-thought reasoning. | Apache-2.0 | No | text-generation, chat, instruction, reasoning | intel/inference-xeon-ibm-granite-granite-3.2-2b-instruct:0.1.0 |
| meta-llama/Llama-3.1-8B-Instruct | 8B-parameter instruction-tuned chat model from Meta's Llama 3.1 family, optimized for multilingual dialogue and tool use. | Llama 3.1 Community License | Yes | text-generation, chat, instruction | intel/inference-xeon-meta-llama-llama-3.1-8b-instruct:0.1.0 |
| meta-llama/Llama-3.2-3B-Instruct | Lightweight 3B instruction-tuned model from Meta's Llama 3.2 family, suited for on-device and latency-sensitive chat use cases. | Llama 3.2 Community License | Yes | text-generation, chat, instruction | intel/inference-xeon-meta-llama-llama-3.2-3b-instruct:0.1.0 |
| microsoft/Phi-4-reasoning | Reasoning-focused variant of Microsoft's Phi-4 model, tuned with chain-of-thought techniques for complex multi-step problem solving. | MIT | No | text-generation, chat, instruction, reasoning | intel/inference-xeon-microsoft-phi-4-reasoning:0.1.0 |
| mistralai/Mistral-7B-Instruct-v0.2 | 7B-parameter instruction-tuned model from Mistral AI, optimized for general-purpose chat and instruction following. | Apache-2.0 | No | text-generation, chat, instruction | intel/inference-xeon-mistralai-mistral-7b-instruct-v0.2:0.1.0 |
| openai/gpt-oss-20b | 20B-parameter open-weight reasoning model from OpenAI's gpt-oss series, tuned for chain-of-thought reasoning and agentic tool use. | Apache-2.0 | No | text-generation, chat, instruction, reasoning | intel/inference-xeon-openai-gpt-oss-20b:0.1.0 |
| Qwen/Qwen3-30B-A3B | Mixture-of-experts language model from Alibaba's Qwen3 series (30B total / 3B active parameters), tuned for efficient reasoning and dialogue. | Apache-2.0 | No | text-generation, chat, instruction, reasoning | intel/inference-xeon-qwen-qwen3-30b-a3b:0.1.0 |
| Qwen/Qwen3-4B | 4B-parameter dense language model from Alibaba's Qwen3 series with dual thinking/non-thinking modes for reasoning and efficient dialogue. | Apache-2.0 | No | text-generation, chat, instruction, reasoning | intel/inference-xeon-qwen-qwen3-4b:0.1.0 |
| Qwen/Qwen3-8B | 8B-parameter dense language model from Alibaba's Qwen3 series with dual thinking/non-thinking modes for reasoning and chat. | Apache-2.0 | No | text-generation, chat, instruction, reasoning | intel/inference-xeon-qwen-qwen3-8b:0.1.0 |
| Qwen/Qwen3-VL-30B-A3B-Instruct | Multimodal vision-language mixture-of-experts model from Qwen3-VL (30B total / 3B active parameters), instruction-tuned for image-and-text chat. | Apache-2.0 | No | text-generation, chat, instruction, multimodal | intel/inference-xeon-qwen-qwen3-vl-30b-a3b-instruct:0.1.0 |
Related pages
- Quickstart — run one of these images and call it
- Prerequisites — Hugging Face token and license approval for gated models
- Supported Intel® platforms — Intel® processors (Intel® Xeon® CPU), and memory sizing per model size
- Deploy on Kubernetes — the four things to change in the manifest for your model
- Security — passing
HF_TOKENfor the gated models above - Supported features — what the multimodal entries accept