GigaChat3-10B-A1.8B is an efficient dialogue model in the GigaChat series. Based on the Mixture-of-Experts (MoE) architecture, it has a total of 10 billion parameters and 1.8 billion active parameters. It adopts the innovative Multi-head Latent Attention (MLA) and Multi-token Prediction (MTP) technologies, aiming to optimize inference throughput and generation speed. The model is trained on 20T tokens of diverse data and supports 10 languages including Chinese, suitable for dialogue scenarios requiring quick responses.

Natural Language ProcessingGgufMultiple Languages

Uni-MoE 2.0-Omni is a fully open-source full-modal model launched by Lychee Technology. It adopts the full-modal 3D RoPE and dynamic-capacity mixture-of-experts architecture, significantly enhancing the language-centered multi-modal understanding, reasoning, and generation capabilities. This version integrates full-modal understanding as well as audio and image generation capabilities.

MultimodalTransformersMultiple Languages

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

Aquif-3.5-Max-42B-A3B is a large language model with 42 billion parameters, which has undergone MXFP4_MOE quantization processing. It optimizes inference efficiency while maintaining high-quality text generation capabilities. This model is based on an advanced mixture-of-experts architecture and is suitable for various natural language processing tasks.

Natural Language ProcessingGguf

This project is the MXFP4_MOE quantized version of the Kanana 1.5-15.7B-A3B instruction model, which can improve the model's operating efficiency in specific scenarios. This model is optimized based on kakaocorp/kanana-1.5-15.7b-a3b-instruct, using a mixture-of-experts architecture and MXFP4 quantization technology.

Natural Language ProcessingGguf

Kimi K2 Thinking is the latest generation of open-source thinking model developed by Moonshot AI. It has powerful deep reasoning ability and tool call function. The model adopts a mixture-of-experts architecture, supports native INT4 quantization, has a 256k context window, and performs excellently in multiple benchmark tests.

Natural Language ProcessingTransformers

Qwen3-VL-30B-A3B-Instruct is the most powerful vision-language model in the Tongyi series. It adopts a mixture-of-experts model architecture and has excellent text understanding and generation capabilities, in-depth visual perception and reasoning capabilities. It supports a 256K long context and video understanding, and can perform inference on a variety of devices.

MultimodalGguf

Cogito v2.1 is a 671 billion parameter mixture-of-experts generative model fine-tuned on instructions. It is released under an open license and supports commercial use. This model can effectively solve complex reasoning and instruction-following problems, and supports over 30 languages and a 128k context length.

Natural Language ProcessingTransformers

This is an MXFP4 Mixture-of-Experts quantized version based on the Intern-S1 model, specifically optimized for image-text-to-text tasks. It improves inference efficiency through quantization technology.

MultimodalGguf

Qwen3-Next-80B-A3B-Thinking-1M-qx64n-mlx is a large language model that adopts a mixture-of-experts architecture and Deckard mixed precision quantization technology. It has 80 billion parameters and a context length of 1M tokens. This model performs excellently in science and technology reasoning and long text processing, showing a 20 - 35% improvement in cognitive benchmark tests compared to instruction models.

Natural Language ProcessingMlx

This is a mixture-of-experts model based on the GLM-4.6 architecture. The REAP method was used to uniformly prune 40% of the experts. The model has 218B parameters and has been converted to a 3-bit quantization version in the MLX format, which is suitable for efficient operation on Apple chip devices.

Natural Language ProcessingTransformersEnglish

OpenModel-1T-A50B-Instruct is a trillion-parameter mixture-of-experts model developed by NeXTHub. It combines the evolutionary thought chain training method to achieve a balance between reasoning ability and energy efficiency, and has deep reasoning ability and 128K long-context processing ability.

Natural Language ProcessingSafetensors

OLGA is an online reinforcement learning positioning agent built on Qwen3-VL-30B-A3B-Instruct, using a mixture-of-experts model with 3.3 billion active parameters. It is trained through a new data recipe that combines existing datasets, new data collection, automatic filtering, and online reinforcement learning, achieving advanced positioning performance among open-source models.

MultimodalTransformersEnglish

Qwen3-VL-32B-Thinking is the most powerful vision-language model in the Qwen series. It has excellent text understanding and generation capabilities, in-depth visual perception and reasoning abilities, long context support, strong spatial and video dynamic understanding abilities, and outstanding agent interaction capabilities. This version adopts a thinking architecture for enhanced reasoning, supporting dense architectures from edge to cloud and mixture-of-experts model architectures.

MultimodalTransformers

Zirel-3 is a specific fine-tuned model based on GLM-4.5-Air-REAP-82B-A12B. It is an 82-billion-parameter mixture-of-experts model compressed using the REAP (Routing Weighted Expert Activation Pruning) technology, significantly reducing the model size while maintaining high performance.

Natural Language ProcessingTransformersEnglish

LFM2-8B-A1B-qx86-hi-mlx is an efficient inference model based on the MLX format, converted from LiquidAI/LFM2-8B-A1B. This model adopts a mixture-of-experts architecture and demonstrates卓越的 efficiency in inference tasks. It is particularly擅长 at complex logical reasoning tasks and supports multilingual processing.

Natural Language ProcessingMlxMultiple Languages

This is an 8-bit quantized version converted from the LiquidAI/LFM2-8B-A1B model, optimized for the Apple MLX framework. This model is a mixture-of-experts model with 8B parameters, supporting multilingual text generation tasks.

Natural Language ProcessingMlxMultiple Languages

This is a mixture-of-experts model based on Qwen3-Coder-30B-A3B-Instruct, with 54 billion parameters and a context length of 1 million. The model has powerful programming capabilities and general scenario processing capabilities through three-step merging and Brainstorm 40X optimization. It is especially integrated with a thinking module, which can perform in-depth reasoning before answering.

Natural Language ProcessingTransformersMultiple Languages

Qwen3-VL is the most powerful vision-language model in the Qwen series. It has excellent text understanding and generation capabilities, in-depth visual perception and reasoning capabilities, long context support, strong spatial and video dynamic understanding capabilities, and outstanding intelligent agent interaction capabilities. This version uses a mixture-of-experts model architecture and supports enhanced reasoning thinking functions.

MultimodalSafetensors

Qwen3-VL is the most powerful vision-language model in the Tongyi large model series, with significant improvements 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 model architecture, supporting deployments of different scales from the edge to the cloud.

MultimodalTransformers