NVIDIA GTC 2026 — Nemotron 3 Super & Omni Added
Added NVIDIA's latest open models from GTC 2026: Nemotron 3 Super (120B MoE, 12B active, 1M context, 7× throughput) and Nemotron 3 Omni (multimodal video/document understanding). Model count now 109. Enhanced community sentiment with badges, stacked sentiment bars, trending indicators, and community stats mined from reviews.
24 New Models Added — Music Gen & 3D Gen Sections Launched
Major expansion: added 24 new model cards across all categories. Model count now 107. New models include Gemma 3n, StarCoder2, MiniMax M2.5, Qwen3.5 32B, Falcon 3, Codestral 25.05, SmolVLM, Molmo, Chatterbox TTS, Dia TTS, MaskGCT, Jina Embeddings v3, Hunyuan Image, Sana, and more. Launched two brand-new sections: Music Generation (ACE-Step, Stable Audio Open, MusicGen) and 3D Generation (TRELLIS 2, Hunyuan3D 2.0, TripoSR). Site now updates continuously.
Missing Models Research — 50+ Models Identified for Addition
Deep research completed across all categories. Key findings: Chinese labs now dominate frontier open-weight models (GLM-5, Kimi K2.5, Qwen 3.5, DeepSeek V3.2). Missing models to add include: Chatterbox TTS (Resemble AI), ACE-Step (music gen), TRELLIS/Hunyuan3D (3D gen), Phi-4-reasoning, SmolVLM, Moonshine Gen 2, Ministral 3, and many more. MoE architecture now dominates frontier models.
GLM-5 744B, Kimi K2.5, GPT-OSS 120B Added
Added latest frontier models: GLM-5 (744B MoE, #1 open-source Quality Index), Kimi K2.5 (Moonshot AI reasoning), GPT-OSS (OpenAI's first open-weight models at 20B and 120B MoE). Updated model count to 107.
Community Analytics Integration
Added real-time model popularity data from HuggingFace — every model card now shows real download counts and community likes.
FLUX.2, Z-Image-Turbo, Wan 2.2 MoE, OLMo 3.1 Added
Added FLUX.2 image generation suite (Klein 4B to Pro 32B), Z-Image-Turbo (sub-second generation), Wan 2.2 with MoE architecture, and OLMo 3.1 (the only fully open-source model family with training data, code, and checkpoints).
Quantization Guide Expanded + Recommended Stacks
Major expansion of the quantization guide with JPEG analogy, real-world scenarios, and "What Breaks First" degradation table. Added 6 complete recommended stacks by hardware tier (RTX 4090, 4070 Ti, 3060, Mac M-series, Air-Gapped, Creative).
Initial Launch — 107 Models Cataloged
First release of the Local AI Model Index. 107 models across 10 categories: LLMs, coding, vision, embeddings, STT, TTS, image generation, and video generation. Search, filter, hardware tiers, and VRAM calculator included.