OpenMMReasoner is a fully transparent two-stage multimodal reasoning solution that covers supervised fine-tuning (SFT) and reinforcement learning (RL). By carefully constructing high-quality datasets, this solution surpasses powerful baseline models in multiple multimodal reasoning benchmark tests, laying a solid empirical foundation for future large-scale multimodal reasoning research.

MultimodalTransformers

Sarashina2.2-Vision-3B is a large Japanese vision-language model trained by SB Intuitions. It is built based on Sarashina2.2-3B-Instruct and the SigLIP image encoder, and has powerful image-to-text conversion capabilities, supporting multimodal processing in Japanese and English.

MultimodalTransformersMultiple Languages

ERNIE-4.5-VL-28B-A3B-Thinking AWQ - INT8 is a multimodal large language model based on Baidu's ERNIE-4.5 architecture. It achieves 8-bit precision through the AWQ quantization technology, significantly reducing memory requirements while maintaining high performance. This model performs excellently in visual reasoning, STEM problem-solving, image analysis, etc., and has powerful multimodal understanding and reasoning abilities.

MultimodalTransformersMultiple Languages

SenseNova-SI is a series of spatial intelligence models built on multimodal foundation models, specifically optimized for spatial understanding capabilities. Through large-scale spatial intelligence data training, it performs excellently in metric estimation, spatial relationship understanding, viewpoint change handling, etc.

MultimodalTransformers

JanusCoder-8B is an open-source code intelligence foundation model built on Qwen3-8B, aiming to establish a unified visual programming interface. This model is trained on JANUSCODE-800K (the largest multimodal code corpus to date) and can handle various visual programming tasks, including data visualization, interactive web UI, and code-driven animation.

MultimodalTransformers

Emu3.5 is a native multimodal model developed by the Beijing Academy of Artificial Intelligence (BAAI). It can jointly predict the next state across vision and language, enabling coherent world modeling and generation. After end-to-end pre-training and large-scale reinforcement learning post-training, it demonstrates excellent performance in multimodal tasks.

MultimodalTransformers

The MLX MXFP4 quantized version of Gemma 3 27B IT QAT is a lightweight open - source multimodal model developed by Google. This model can simultaneously process text and image inputs and generate text outputs. It has a large context window of 128K and supports over 140 languages, making it suitable for various text generation and image understanding tasks.

MultimodalMlx

Qwen3-VL is the most powerful vision-language model in the Tongyi large model series, with comprehensive upgrades in text understanding, visual perception, spatial understanding, video processing, etc., providing excellent multimodal capabilities.

MultimodalSafetensors

The Ming-flash-omni preview version is a multimodal large model built on the Ling-Flash-2.0 sparse mixture of experts (MoE) architecture, with a total of 100B parameters and only 6B parameters activated per token. This model is a comprehensive upgrade based on Ming-Omni, showing significant improvements in multimodal understanding and generation, especially in speech recognition, image generation, and segmentation editing.

MultimodalDiffusersEnglish

Qwen3-VL-8B-Thinking is an 8-billion-parameter multimodal large language model developed by the Qwen team at Alibaba Cloud. It is designed for deep multimodal reasoning, supporting visual understanding, long context processing, and structured chain-of-thought generation, and performs excellently in complex reasoning tasks.

MultimodalMlx

Qwen3-VL-4B-Instruct is a 4-billion-parameter instruction-tuned multimodal large language model launched by the Alibaba Cloud Qwen team. It is optimized for Qualcomm NPUs, integrating powerful visual-language understanding capabilities with dialogue fine-tuning functions, and is suitable for practical application scenarios such as chat reasoning, document analysis, and visual dialogue.

MultimodalGguf

Bee-8B is an advanced fully open-source multimodal large language model that focuses on data quality and aims to narrow the performance gap with proprietary models. By using the high-quality Honey-Data-15M corpus and the advanced data processing pipeline HoneyPipe, it demonstrates excellent performance in complex reasoning and other aspects.

MultimodalTransformersEnglish

Apriel-1.5-15b-Thinker is a multimodal reasoning model developed by the ServiceNow SLAM Lab, with a parameter scale of 15 billion. It has powerful text and image understanding and reasoning capabilities, adopting a chained reasoning mode of 'think-answer', aiming to achieve performance comparable to large models with a smaller model scale.

MultimodalTransformers

Apriel-1.5-15b-Thinker is a 15-billion parameter multimodal reasoning model developed by ServiceNow. It has the ability to reason with text and images, and its performance can rival models 10 times larger. It scored 52 points on the manual analysis index and performed excellently in enterprise domain benchmark tests.

MultimodalTransformers

PaDT Pro 3B is a multimodal large language model based on Qwen2.5VL-3B. It adopts a unified paradigm of decodable block tokens and can directly generate text and visual outputs, achieving advanced performance in multiple visual perception and understanding tasks.

MultimodalTransformersMultiple Languages

Ming-UniVision is a large multimodal language model that first integrates continuous visual representations into the next-token prediction framework, unifying vision and language under a single autoregressive paradigm without discrete quantization or modality-specific heads. This model supports joint image understanding and generation, converges faster in vision-language training, and also supports multi-round context visual tasks.

MultimodalTransformersMultiple Languages

This is a professional quantized version of the Qwen3-Omni model with 31.7B parameters. It uses advanced INT8+FP16 mixed-precision quantization technology, reducing memory usage by over 50%. It supports intelligent GPU/CPU hybrid inference, enabling large multimodal models to run efficiently on consumer-grade hardware.

MultimodalSafetensorsMultiple Languages

Apriel-1.5-15b-Thinker is a multimodal reasoning model developed by ServiceNow, with 15 billion parameters. It performs excellently in text and image reasoning tasks and can compete with models 10 times larger than it.

MultimodalTransformers

Lapa LLM 12B PT is an open-source large language model developed based on Google Gemma 3-12B, specifically optimized for Ukrainian language processing. It is developed by research teams from multiple Ukrainian universities and performs excellently in Ukrainian language processing, with efficient instruction fine-tuning and multimodal capabilities.

MultimodalTransformersOther

UniPixel-3B is a unified multimodal large language model for pixel-level vision-language understanding, which can flexibly support various fine-grained tasks, including image/video segmentation, region understanding, and the novel PixelQA task. The model jointly requires object-centric reference, segmentation, and question answering in videos, achieving pixel-level visual reasoning capabilities.

MultimodalSafetensors