SAM 3 is the third-generation promptable segmentation foundation model launched by Meta, which uniformly supports image and video segmentation tasks. Compared with its predecessor SAM 2, it introduces the ability of open vocabulary concept segmentation, can handle a large number of text prompts, and achieves 75-80% of human performance in the SA-CO benchmark test.

Computer VisionTransformersEnglish

Qwen3-VL-2B-Instruct-FP8 is the FP8 quantized version of the most powerful vision-language model in the Qwen series. It uses fine-grained fp8 quantization with a block size of 128, and its performance is almost the same as that of the original BF16 model. This model has excellent text understanding and generation capabilities, in-depth visual perception and reasoning abilities, long context support, and enhanced spatial and video dynamic understanding abilities.

MultimodalTransformers

Qwen3-VL-32B-Instruct-FP8 is the FP8 quantized version of the most powerful vision-language model in the Qwen series. It uses fine-grained fp8 quantization with a block size of 128, and its performance metrics are almost the same as the original BF16 model. This model has excellent text understanding and generation capabilities, in-depth visual perception and reasoning abilities, long context support, and strong spatial and video dynamic understanding capabilities.

MultimodalTransformers

Qwen3-VL-32B-Thinking-FP8 is the FP8 quantized version of the most powerful vision-language model in the Qwen series. It uses fine-grained fp8 quantization technology with a block size of 128, and its performance indicators are almost the same as those of the original BF16 model. This model has excellent text understanding and generation capabilities, in-depth visual perception and reasoning abilities, long context support, and powerful agent interaction capabilities.

MultimodalSafetensors

Apertus is a fully open multilingual language model with a parameter scale of 70B and 8B. It supports over 1000 languages and long context processing. It only uses fully compliant and open training data, and its performance can rival that of closed-source models.

Natural Language ProcessingTransformersOther

Apertus is a fully open multi-language language model with a parameter scale of 70B and 8B. It supports over 1000 languages and long contexts, and only uses fully compliant and open training data. Its performance can be comparable to that of models trained with closed-source data.

Natural Language ProcessingTransformersOther

Apertus is a fully open multilingual large language model developed by Swiss AI, offering two parameter scales of 7 billion and 8 billion. This model supports over 1000 languages, uses fully compliant and open training data, and its performance can rival that of closed-source models. Apertus was pre-trained on 15T tokens and adopts a phased curriculum training method, supporting a context length of up to 65,536 tokens.

Natural Language ProcessingTransformersOther

Apertus is a fully open multilingual large language model developed by Swiss AI. It has parameter scales of 8 billion and 70 billion, supports over 1000 languages and long context processing, and only uses fully compliant open training data. Its performance can rival that of closed-source models.

Natural Language ProcessingTransformersOther

Qwen3-VL-30B-A3B-Thinking-FP8 is the FP8 quantized version of the most powerful vision-language model in the Tongyi series. It uses fine-grained fp8 quantization with a block size of 128, and its performance indicators are almost the same as those of the original BF16 model. This model has excellent text understanding and generation capabilities, in-depth visual perception and reasoning abilities, and supports long contexts and video understanding.

MultimodalSafetensors

Qwen3-VL-30B-A3B-Instruct-FP8 is the FP8 quantized version of the most powerful vision-language model in the Tongyi series. It uses fine-grained FP8 quantization with a block size of 128, and its performance is almost the same as that of the original BF16 model. This model has excellent text understanding and generation capabilities, in-depth visual perception and reasoning capabilities, long context support, and strong agent interaction capabilities.

MultimodalSafetensors

Qwen3-VL-235B-A22B-Thinking-FP8 is the FP8 quantized version of the most powerful vision-language model in the Tongyi series. It uses fine-grained fp8 quantization with a block size of 128, and its performance metrics are almost the same as those of the original BF16 model. This model has excellent text understanding and generation capabilities, in-depth visual perception and reasoning capabilities, long context support, and strong agent interaction capabilities.

MultimodalSafetensors

Qwen3-VL-235B-A22B-Instruct-FP8 is the FP8 quantized version of the most powerful vision-language model in the Tongyi series. It uses fine-grained fp8 quantization with a block size of 128, and its performance is almost the same as that of the original BF16 model. This model has excellent text understanding and generation capabilities, in-depth visual perception and reasoning abilities, long context support, and strong spatial and video dynamic understanding abilities.

MultimodalSafetensors

VideoChat-R1_5-7B is a video-text interaction model built on Qwen2.5-VL-7B-Instruct, supporting multimodal tasks and excelling in video Q&A functions. This model enhances spatio-temporal perception ability through reinforcement fine-tuning and adopts an iterative perception mechanism to strengthen multimodal reasoning.

MultimodalTransformersEnglish

The monthly iteration version of the Tongyi Image Editing Model, which supports multi-image editing, improves the consistency of single-image editing, and natively supports ControlNet, bringing more powerful functions and better effects to image editing.

MultimodalDiffusersMultiple Languages

Borealis is the first automatic speech recognition (ASR) audio large language model for Russian, trained on approximately 7,000 hours of Russian audio data. This model supports the recognition of punctuation marks in audio. Its architecture is inspired by Voxtral but has been improved, and it performs excellently in multiple Russian ASR benchmark tests.

Audio ProcessingPytorchOther

Cogito v2 is a generative hybrid inference model fine-tuned by instructions. It uses iterative distillation and amplification training methods and performs excellently in coding, STEM, instruction following, etc. It supports over 30 languages and a 128k context length.

Natural Language ProcessingTransformers

Cogito v2 is a generative large language model based on instruction tuning, with hybrid reasoning ability, supporting a 128k context length and multilingual processing. This model adopts the Iterative Distillation and Amplification (IDA) training strategy and performs excellently in tasks such as coding, STEM, and instruction following.

Natural Language ProcessingTransformers

Cogito v2 is an instruction-tuned generative model based on Llama-3.1-70B. It has hybrid inference capabilities, supports multiple languages and long contexts, and performs excellently in areas such as coding and STEM. It adopts an iterative distillation and amplification training strategy and can be used for commercial purposes.

Natural Language ProcessingTransformers

Llama-3.3-Nemotron-Super-49B-v1.5 is an efficient large language model developed by NVIDIA, derived from Meta Llama-3.3-70B-Instruct. This model performs excellently in inference, chat interaction, and agent tasks. It significantly reduces memory usage through neural architecture search technology and supports a context length of 128K tokens. Its capabilities in multiple aspects such as mathematics, code, science, and tool invocation have been enhanced.

Natural Language ProcessingTransformersEnglish

FastwebMIIA is a 7-billion-parameter Italian artificial intelligence model developed by Fastweb, designed specifically for Italian and English, and supports long document processing.

Natural Language ProcessingTransformersMultiple Languages