LLaVA-OneVision 1.5
Fully open multimodal (images/video+text) models & training stack; strong results and reproducible recipes.
Sorting
Gemma 2
Google’s open-weight 9B/27B models; practical to run locally and widely adopted for fine-tuning and apps.
InternVL 2.5
Open multimodal family (1B–78B); 78B surpasses 70% on MMMU; broad image/video understanding.
Mixtral 8x22B (Mistral)
Sparse MoE open model delivering top cost/performance among community LLMs; widely fine-tuned and quantized.
MiniCPM-V 2.6
Edge-friendly multimodal 8B model (images/video) with quantized variants for low-VRAM local inference.
DeepSeek-R1 (and distilled checkpoints)
Open (MIT-licensed) reasoning model with distilled 1.5B–70B local checkpoints; known for chain-of-thought quality.
Falcon 2 11B
TII’s newer open series (text+VLM variants) optimized for efficient inference; Apache-style release.