Optimum Intel 2.1 exports six new AI architectures

Hugging Face’s Optimum Intel 2.1 adds export and inference support for six model architectures: Gemma 3n, Gemma 4 Unified, Qwen3-Omni-MoE, Qwen3-VL-Embedding, SmolLM3, and FLUX.2. The release announcement says the package was co-developed and validated with three companion releases—OpenVINO 2026.3, OpenVINO GenAI 2026.3, and NNCF 3.3. For builders already testing local inference as a control surface, the useful fact is broader multimodal coverage in one Intel deployment path; pinning those three validated dependency versions is the conservative production move.