INTELLECT-3 is a Mixture of Experts (MoE) inference model developed by Prime Intellect with a parameter scale of 106B (12B active parameters). It is supervised and fine-tuned based on GLM-4.5-Air-Base and trained through large-scale reinforcement learning, showing excellent performance in mathematics, coding, and inference tasks.

Natural Language ProcessingGguf

INTELLECT-3 is a Mixture of Experts (MoE) model with 106 billion parameters, trained through large-scale reinforcement learning. It demonstrates excellent performance in mathematics, coding, and reasoning benchmark tests. The model, training framework, and environment are all open-sourced under a permissive license.

Natural Language ProcessingTransformersEnglish

This is the quantized version of the ai-sage/GigaChat3-10B-A1.8B model, using the GGUF format and requiring support from a specific llama.cpp branch. The model adopts a Mixture of Experts architecture, with a total of approximately 11.8 billion parameters, of which approximately 1.8 billion are activated parameters.

Natural Language ProcessingGgufMultiple Languages

GLM-4.5-Iceblink-v2-106B-A12B-FP8 is a version based on the GLM-4.5-Iceblink-v2-106B-A12B model, using the most advanced mixture of experts quantization method for FP8 quantization. This model is specifically optimized for Ada, Hopper, or Blackwell series GPUs that support hardware FP8, significantly improving inference efficiency while maintaining high-quality output.

Natural Language ProcessingSafetensors

GigaChat3-10B-A1.8B is a dialogue model in the GigaChat series, based on the Mixture of Experts (MoE) architecture, with a total of 10 billion parameters, of which 1.8 billion are active parameters. This model uses multi-head latent attention and multi-token prediction technologies, supports a long context of 256,000 tokens, and performs excellently in multilingual dialogue and reasoning tasks.

Natural Language ProcessingSafetensorsMultiple Languages

GigaChat3-10B-A1.8B-base is the basic pre-trained model of the GigaChat series, adopting the Mixture of Experts (MoE) architecture with a total of 10 billion parameters and 1.8 billion active parameters. The model integrates Multi-Head Latent Attention (MLA) and Multi-Token Prediction (MTP) technologies, and has the advantage of high throughput during inference.

Natural Language ProcessingSafetensorsMultiple Languages

This is a large language model with 172B parameters obtained by uniformly pruning 25% of the experts in MiniMax-M2 using the REAP method. It has been optimized and quantized specifically for llama.cpp, supporting multiple quantization levels and can run in LM Studio or projects based on llama.cpp.

Natural Language ProcessingGgufEnglish

This is a large language model with 139B parameters obtained by uniformly pruning 40% of the experts in MiniMax-M2 using the REAP method. It adopts the GLM architecture and Mixture of Experts (MoE) technology, and undergoes various quantization processes through llama.cpp, suitable for text generation tasks.

Natural Language ProcessingGgufEnglish

Hunyuan OCR is an end-to-end OCR expert VLM model driven by the Hunyuan native multimodal architecture. With a lightweight design of only 1 billion parameters, it has achieved state-of-the-art results in multiple industry benchmark tests. This model excels in handling complex multilingual document parsing and performs well in practical application scenarios such as text localization, open-domain information extraction, video subtitle extraction, and image translation.

MultimodalTransformersMultiple Languages

This is the MXFP4_MOE quantized version of the MiniMax-M2-REAP-172B-A10B model, which is a memory-efficient compressed model. Through the REAP (Routing-Weighted Expert Activation Pruning) method, the model is compressed from 230B parameters to 172B parameters while maintaining performance, with a 25% reduction in size. It is suitable for resource-constrained environments, local deployment, and academic research.

Natural Language ProcessingGguf

MiniMax-M2-REAP-162B-A10B is an efficiently compressed version of MiniMax-M2. It uses the REAP (Routing Weighted Expert Activation Pruning) method to reduce the model size by 30% while maintaining almost the same performance, compressing from 230B parameters to 162B parameters, significantly reducing the memory requirement.

Natural Language ProcessingTransformersEnglish

MiniMax-M2-REAP-172B-A10B is a memory-efficient compressed variant of MiniMax-M2. It uses the REAP expert pruning method, reducing the model size by 25% from 230B parameters to 172B parameters while maintaining almost the same performance.

Natural Language ProcessingTransformersEnglish

This model is a GGUF format model generated by in-depth fine-tuning of the Qwen3-Coder-30B-A3B-Instruct Mixture of Experts model using Brainstorm 20x technology. It is designed for programming and general purposes, supports a 256K long context, and is optimized for code generation and understanding of mainstream and niche programming languages.

Natural Language ProcessingGgufMultiple Languages

Moondream 3 Preview HF is a re-implementation of the Moondream 3 (Preview) model based on the HuggingFace Transformers architecture specification, making it fully compatible with the Hugging Face ecosystem. It is a multimodal vision-language model that uses a Mixture of Experts (MoE) text backbone, with approximately 9 billion parameters and 2 billion active parameters.

MultimodalTransformers

This is the MXFP4_MOE quantized version of the MiniMax-M2-THRIFT model. It has been compressed based on the original model, including 25% expert pruning (from 256 to 192) and setting top_k=8. Meanwhile, it retains the characteristics of the encoding model and can be used for text generation tasks.

Natural Language ProcessingGguf

Qwen3-Coder-REAP-363B-A35B is a sparse mixture-of-experts model obtained by pruning 25% of the experts from Qwen3-Coder-480B-A35B-Instruct using the REAP method. While maintaining performance close to the original model, it significantly reduces the parameter scale and memory requirements, making it particularly suitable for code generation and intelligent coding scenarios with limited resources.

Natural Language ProcessingTransformersEnglish

Qwen3-VL is the most powerful vision - language model in the Tongyi series, with comprehensive upgrades in text understanding and generation, visual perception and reasoning, context length, spatial and video dynamic understanding, and agent interaction capabilities. This model offers both dense architecture and mixture - of - experts architecture, supporting flexible deployment from edge devices to the cloud.

MultimodalTransformers

MiniMax-M2 is a small Mixture of Experts (MoE) model built specifically for maximizing coding and agent workflows. It has a total of 230 billion parameters, with 1 billion active parameters. This model excels in coding and agent tasks while maintaining strong general intelligence. It is compact, fast, and cost-effective.

Natural Language ProcessingTransformers

Qwen3-VL is the most powerful vision-language model in the Tongyi series. It adopts a Mixture of Experts (MoE) architecture and provides weights in the GGUF format, supporting efficient inference on devices such as CPUs and GPUs. The model has been comprehensively upgraded in text understanding, visual perception, spatial understanding, video processing, etc.

MultimodalGguf

Qwen3-VL is the most powerful vision-language model in the Qwen series to date. It has been comprehensively upgraded in text understanding and generation, visual perception and reasoning, context length, spatial and video dynamic understanding, and agent interaction capabilities. This model adopts a Mixture of Experts (MoE) architecture, providing excellent multimodal processing capabilities.

MultimodalGguf