LLM / Multimodal Text
| Model | Params | License | Quant types | CUDA | ROCm | Intel Arc | Vulkan | Status |
|---|
| Qwen3.6-27B | 27B dense | Apache-2.0 | FP8 · BF16 · int4 | ✓ FP8 / AWQ | ✓ vLLM-ROCm int4 | △ IPEX-LLM int4 (FP8 emul) | △ llama.cpp Q4 | Live |
| Qwen3-Omni-30B-A3B Instruct | 30B MoE / 3B act | Apache-2.0 | AWQ-int4 · FP8 · BF16 | ✓ AWQ-int4 | △ AWQ on RDNA3 | △ llm-scaler-vLLM | ✗ no MoE/omni | Live |
| Qwen2.5-VL-7B | 7B | Apache-2.0 | int8 · BF16 · GGUF | ✓ int8 | ✓ ROCm | △ IPEX-LLM (Qwen-VL) | △ llama.cpp (text-only) | Live |
Video Generation
| Model | Params | License | Quant types | CUDA | ROCm | Intel Arc | Vulkan | Status |
|---|
| Wan2.2 T2V-A14B | 14B MoE | Apache-2.0 | NVFP4 · FP8 · BF16 | ✓ NVFP4 (Blackwell) | ✗ NVFP4 CUDA-only | ✗ NVFP4 CUDA-only | ✗ | Live |
| Wan2.2 I2V-A14B | 14B MoE | Apache-2.0 | NVFP4 (shares T2V enc/VAE) | ✓ NVFP4 | ✗ | ✗ | ✗ | Live |
| Wan2.2 S2V-14B | 14B | Apache-2.0 | FP8 · BF16 | ✓ FP8 | ✗ | ✗ FP8 not native | ✗ | Abandoned |
| LTX-2 19B (audio+video) | 19B | LTX-2 Community (commercial <$10M rev) | FP8 distilled | ✓ FP8 | ? untested | ✗ FP8 not native | ✗ | Rollback |
| LTX-2.3 22B (audio+video) | 22B | LTX-2 Community | fp8 self-quant · BF16 | ✓ fp8 self-quant | ? untested | ✗ FP8 not native | ✗ | Live |
Image Generation
| Model | Params | License | Quant types | CUDA | ROCm | Intel Arc | Vulkan | Status |
|---|
| Qwen-Image-2512 (T2I) | 20B MMDiT | Apache-2.0 | bnb int8 · 4-bit · BF16 | ✓ bnb int8 | △ likely (bf16) | △ diffusers-XPU bf16/NF4 | ✗ diffusers (no Vulkan) | Live |
| Qwen-Image-Edit-2511 | 20B MMDiT | Apache-2.0 | GGUF Q6_K · 4-bit · 8-bit | ✓ GGUF/bf16 | ✓ ROCm GGUF | △ GGUF→Vulkan (SYCL slow) | ✓ llama.cpp Vulkan | Live |
Text-to-Speech (TTS)
| Model | Params | License | Quant types | CUDA | ROCm | Intel Arc | Vulkan | Status |
|---|
| VibeVoice-7B ★ | 7B | MIT (community mirror) | bnb 8-bit · BF16 | ✓ int8 | ✓ SDPA | △ IPEX bf16/NF4 | ✗ | Live |
| VibeVoice-1.5B | 1.5B | MIT (MS: research-only note) | BF16 | ✓ | ✓ SDPA | △ IPEX bf16 | ✗ | Eval |
| Chatterbox | 0.5B | MIT | BF16 | ✓ | ✓ | △ IPEX bf16 | ✗ | Eval |
| Kani-TTS-2-en | ~0.45B | Apache-2.0 | BF16 | ✓ | △ NeMo codec | ✗ NeMo codec | ✗ | Eval |
| LuxTTS | ZipVoice | Apache-2.0 | BF16 (48 kHz) | ✓ | ? native deps | ? native deps | ✗ | Eval |
| Qwen3-TTS-1.7B ★ | 1.7B ×2 | Apache-2.0 | BF16 (Base+VoiceDesign) | ✓ | △ likely | △ IPEX bf16 | ✗ | Live |
| Dia-1.6B | 1.6B | Apache-2.0 | BF16 | ✓ | — | — | — | Retired |
✓ stack supports it △ partial/likely ? untested ✗ unsupported · ★ = current TTS pick.
Columns are software/quant-stack support only — per-card hardware is on Cards, live deployments on Services, measured speed on Benchmarks.