Model families
- Llama 1/2/3
- IQuest-Coder V1
- IQuest-Coder Loop (two-pass looped Llama, gated global/local attention)
- Llama 4 (MoE)
- Llama 4 VLM
- Llama 3.2 Vision (tiled ViT + gated cross-attention)
- Qwen 2 / 2.5
- Qwen 3
- Qwen 3 MoE
- Qwen 3 Next (Attention + GatedDeltaNet + MoE)
- Qwen 3.5 / 3.8 (Attention + GatedDeltaNet hybrid)
- Qwen 3.5 / 3.8 VLM
- Qwen 3.5 / 3.6 MoE (hybrid)
- Qwen 3.5 / 3.6 MoE VLM
- Qwen3.8-Flash-Next VLM (QSA + GatedDeltaNet + n-gram PLE + MoE)
- Gemma 1
- Gemma 2
- Gemma 3
- Gemma 4
- DiffusionGemma (block-diffusion, Gemma 4 MoE backbone)
- LLaDA-2 MoE (masked-diffusion LM, DeepSeek-V3-style MoE)
- Gemma 3 VLM
- Gemma 4 VLM
- Gemma 4 Unified (encoder-free text + vision + audio)
- LLaVA (CLIP/SigLIP + Llama/Qwen2)
- Granite Vision (SigLIP + Granite)
- Granite 4 Vision (SigLIP + Granite 4 hybrid)
- DeepSeek-OCR (SAM + CLIP + DeepSeek MoE)
- DeepSeek-OCR 2 (SAM + Qwen2 resampler + DeepSeek MoE)
- Unlimited-OCR (SAM + CLIP + DeepSeek MoE, ring sliding decode)
- DeepSeek-VL2 (SigLIP + downsample MLP + DeepSeek-V2 MoE)
- LLaVA-Bunny (SigLIP + Qwen2)
- FastVLM (FastViTHD + Qwen2)
- ERNIE 4.5 MoE VL (DFNRope + MoE)
- Hunyuan-VL (ViT + XD-RoPE)
- Aya Vision (SigLIP + Cohere2)
- PaliGemma (SigLIP + Gemma)
- Pixtral (2D-RoPE ViT + Mistral)
- Mistral 3 VLM (Pixtral ViT + Mistral)
- Qwen2-VL
- Qwen2.5-VL
- Qwen3-VL
- Qwen3-VL MoE
- Qwen3-Omni MoE (thinker)
- Cohere Compass / North-Micro-Vision
- PaddleOCR-VL
- dots.ocr (dots_vit + Qwen2)
- Falcon-OCR (early fusion)
- Jina VLM (SigLIP-so400m + 2x2 attention pooling + Qwen2 text)
- GLM-4V
- GLM-4V MoE
- GLM-OCR
- Youtu-LLM (DeepSeek-V2-style MLA decoder, text-only)
- Youtu-VL (SigLIP2 windowed-attn + DeepSeek-V3 MLA)
- InternVL (InternViT + pixel-shuffle + Qwen2)
- LLM-jp VL (SigLIP2 + pixel-shuffle mlp1 + Llama/Qwen3)
- Mage-VL (Mage-ViT 3D-RoPE + 2x2 merger + Qwen3)
- GOT-OCR 2.0 (SAM ViT-B + linear projector + Qwen2-0.5B)
- LocateAnything (MoonViT + MLP connector + Qwen2, grounding)
- SmolVLM (SigLIP + pixel-shuffle + SmolLM2)
- Idefics2 (SigLIP + perceiver resampler + Mistral)
- MiniCPM-o (dynamic SigLIP + resampler + Qwen3-VL text)
- MiniCPM-V 4.6 (SigLIP + VitMerger + Merger + Qwen3.5 text)
- Moondream 3 (custom ViT + custom decoder)
- Moondream 2 (SigLIP-style ViT + Phi text)
- Florence-2 (DaViT + BART seq2seq, task prompts)
- Nemotron-Parse (C-RADIO ViT-H + mBART seq2seq OCR)
- Gemma 3n
- Gemma 3n VLM (MobileNetV5 + Gemma3n)
- Phi 1 / 2
- Phixtral (Phi-2 backbone + sparse MoE)
- Phi 3
- Phi-4 Multimodal (SigLIP2 NaFlex + Phi4)
- Phi-4 SigLIP Vision (SigLIP2 NaFlex + Phi3 text)
- Phi 3.5 Vision (CLIP + Phi3)
- Molmo (CLIP ViT + OLMo-style text)
- Molmo 2 (custom ViT + Molmo2 text)
- Molmo-Point (point prediction + Molmo2 text)
- Phi 3 Small
- Phi MoE
- gpt-oss (MoE)
- MiniMax-M2 (MoE, 256 experts)
- MiniMax-M3 (hybrid dense/MoE, block-sparse attention)
- MiniMax-M3-VL (CLIP ViT + M3 hybrid dense/MoE)
- Muse Glimmer 30B (text-only)
- Muse Glimmer 30B VLM
- Mixtral (MoE)
- Qwen 2 MoE
- OLMoE (MoE)
- Databricks DBRX (MoE)
- DeepSeek v1
- DeepSeek v2
- DeepSeek v3 / R1
- DeepSeek v3.2
- DeepSeek v4
- dots.llm1 (MoE)
- Command R (Cohere)
- Command R+ (Cohere2)
- Command MoE (Cohere2 MoE)
- InternLM 2
- InternLM 3
- Baichuan
- GLM 4
- GLM 4 MoE
- GLM 4 MoE Lite
- GLM MoE DSA
- ERNIE 4.5
- ERNIE 4.5 MoE
- Hunyuan MoE
- Hunyuan v1 Dense
- MiMo (multi-token prediction)
- Xiaomi MiMo v2 Flash (hybrid sliding/full attention MoE)
- Ling / Bailing MoE (shared + routed experts)
- Ling / Ring linear-attention MoE (GLA + full attention hybrid)
- Arcee AFMoE / Trinity (sliding + full hybrid MoE)
- Klear MoE (coefficient-blended shared expert)
- Apertus (dense)
- Seed-OSS
- Granite (dense)
- BitNet b1.58 (ternary)
- ExaOne 3
- ExaOne 4
- ExaOne MoE
- Solar Open
- OLMo 1
- OLMo 2
- OLMo 3
- Apple OpenELM (layer-wise scaling)
- GPT-2 (learned absolute positions, Conv1D weights)
- GPT-BigCode (StarCoder / SantaCoder, multi-query attention)
- GPT-NeoX (EleutherAI Pythia, partial RoPE, parallel residual)
- nanochat (weightless RMSNorm, relu^2 MLP, softcapped logits)
- StarCoder 2
- Mellum 2 (JetBrains code)
- Laguna XS / S (Poolside code, hybrid MoE)
- Kyutai Helium (dense Llama shape, traditional RoPE)
- TeleAI TeleChat3 (dense Llama shape, YaRN RoPE)
- MiniCPM 1
- MiniCPM 3
- StableLM
- SmolLM 3
- Ministral 3
- Mistral 3
- Mistral 4 (MLA)
- Nemotron-4
- Mamba 1 / Falcon Mamba
- Mamba 2
- Jamba (Mamba + Transformer + MoE)
- Nemotron-H (Mamba2 + Attention + MLP/MoE hybrid)
- Nemotron-H Nano Omni VLM
- Nemotron-NAS
- Falcon-H1 (Mamba2 + Attention parallel hybrid)
- LFM2 (short-conv + attention hybrid)
- LFM2-MoE (sigmoid-gated experts)
- LFM2-VL (packed-patch ViT + LFM2 hybrid text)
- Inkling (relative-bias attention + shared-expert MoE)
- Inkling VLM (HMLP vision + relative-bias MoE text)
- PLaMo 2 (Mamba + attention hybrid)
- Granite 4 (Mamba2 + attention hybrid)
- Kimi Linear (MLA + GatedDeltaNet hybrid)
- Kimi K3 (KDA + gated NoPE-MLA + latent SiTU MoE + AttnRes)
- Kimi K3 VLM (MoonViT3D + K3 latent MoE)
- Kimi-VL (MoonViT + DeepSeek-V3 MoE)
- Kimi-VL 2.5 (MoonViT + DeepSeek-V3 MoE)
- LongCat Flash (MLA + MoE, dual sublayer)
- LongCat Flash + N-gram embedding
- Step-3.5 (Sigmoid MoE gate + SwitchGLU)
- Step-3.7 (ViT + Step-3.5 MoE)
- RWKV v7
- RecurrentGemma (Griffin: RGLRU + attention)
- Whisper (encoder-decoder ASR)
- Kokoro (StyleTTS2 + iSTFTNet)
- Nemotron VoiceChat (FastConformer + Nemotron-H + EAR-TTS + codec, full duplex)
- BERT / MiniLM encoder
- XLM-RoBERTa encoder
- ModernBERT encoder
- SigLIP text tower
- EmbeddingGemma (bidirectional Gemma 3)
- EmbeddingGemma 2 (multimodal bidirectional Gemma 4)
- Qwen3-Embedding (last-token)
- Qwen3-VL-Embedding (multimodal)
- LFM2.5-Embedding (bidirectional LFM2)
- Nemotron-3-Embed (bidirectional Ministral 3)
- Bidirectional Llama / LLM2Vec embedder
- Llama-Nemotron-VL-Embed (multimodal)
- ColIdefics3 (late-interaction, multimodal)
- ColQwen2.5 (late-interaction, multimodal)
- Cross-encoder sequence classifier (BERT / XLM-RoBERTa / ModernBERT)
- RT-DETRv2
- PP-DocLayoutV3
- RF-DETR
- DINOv2
- SAM 3D Body
- SAM 3D Objects
- MoGe3
- Sapiens2
- Video Depth Anything
- SAM 3
- SAM 3.1
- FLUX.2 Klein
- Z-Image
- Prism Bonsai (ternary FLUX.2 Klein)
- ERNIE-Image
- Mage-Flow
- Qwen-Image
- Ideogram 4
- Ming-Image
- LLaDA-Image