Zhipu AI/Text / LLMMajor release
Zhipu AI Releases MIT-Licensed GLM-5.2 MoE Model
The new bilingual model from the Chinese AI firm uses a Mixture of Experts architecture and sparse attention under a fully permissive license.
The feed
Latest releases
Every new open-source model release and major update — aggregated from across the ecosystem, deduplicated, and refreshed every 12 hours.
164 releases
Weibo AI/Reasoning
Weibo AI Releases VibeThinker-3B, a Compact Reasoning Model
The new 3-billion-parameter model from the Chinese tech giant focuses on challenging benchmarks in mathematics, coding, and graduate-level questions.
Moonshot AI/CodeMajor release
Moonshot AI Releases Kimi, a Multimodal Coding Model
The new Mixture-of-Experts model from the Chinese AI company can generate code while also understanding visual inputs, a rare combination in open models.
Zyphra/Text → Speech
Zyphra Releases Open-Source Zonos 2 TTS Model
The new text-to-speech model offers a commercially permissive alternative for developers in a field still dominated by closed-source APIs.
Google DeepMind/Text / LLM
Google Releases Open-Source DiffusionGemma 26B Model
The new 26B parameter model from DeepMind uses a diffusion-based architecture, a technique more common in image generation, to produce text.
Zhipu AI/Image → Video
Zhipu AI Releases SCAIL-2 for Character Animation
The new open-source diffusion model from the company's research arm generates video clips from a single character image and a sequence of poses.
Cohere/Code
Cohere Releases North-Mini-Code, an Open MoE Model
The new Apache 2.0-licensed model is designed for code generation and agentic chat applications, using a Mixture-of-Experts architecture for efficiency.
Boson AI/Text → Speech
Boson AI's Higgs Audio v3 Offers Expressive, Multilingual TTS
The new 4-billion-parameter text-to-speech model is available for non-commercial use, promising fine-grained control over vocal delivery.
MiniMax/Vision-LanguageMajor release
MiniMax Releases M3, a Multimodal MoE Model
The new open-weight model from MiniMax AI combines vision, coding, and reasoning using a Mixture-of-Experts architecture.
Ideogram/Text → Image
Ideogram 4.0: A 9.3B Open-Weight Text-to-Image Model
The new 9.3 billion parameter model uses a Diffusion Transformer architecture and excels at rendering coherent text within generated images.
Google DeepMind/Any-to-AnyMajor release
Google Releases Gemma 4, a 12B 'Any-to-Any' Model
The new 12-billion-parameter model from Google DeepMind is designed to handle a flexible mix of data types, moving beyond traditional text and image inputs.
NVIDIA/Image → Video
NVIDIA Releases SANA, a Camera-Controllable Video Model
The new model, SANA-WM, uses a bidirectional diffusion process to give creators fine-grained control over camera movement and video editing.
NVIDIA/Speech → Text
NVIDIA Releases Nemotron-3.5 Streaming ASR Model
The 600-million-parameter model uses a FastConformer architecture for real-time, multilingual speech-to-text applications.
ByteDance/Any-to-AnyMajor release
ByteDance Releases Lance, a Unified Generative AI Model
The 3-billion-parameter model handles image and video generation, editing, and understanding from a single set of weights under a permissive license.
SenseTime/Any-to-Any
SenseTime Releases 8B 'Any-to-Any' Infographic Model
The new 8B-parameter SenseNova U1 model from SenseTime is designed for complex multimodal tasks, including the in-conversation generation and editing of infographics.
Lightricks/Image → Video
Lightricks Releases LoRA for AI Lip-Dubbing
The new 'Identity-Control' adapter fine-tunes the company's LTX-2.3 video model to create realistic lip-syncing for dubbing workflows.
Tencent/Text / LLM
Tencent Releases 1.8B Model for Multilingual Translation
The 1.8 billion-parameter model from the Chinese tech giant is designed for high-quality translation across a wide range of language pairs.
Supertone/Text → Speech
Supertone Releases On-Device Multilingual TTS Model
The new Supertonic 3 model supports seven languages and is optimized for local inference with the portable ONNX format.
NVIDIA/Image Editing
NVIDIA Releases PiD for High-Quality Image Upscaling
The new component is a specialized VAE decoder that works with Stability AI's Z-Image model to enhance super-resolution tasks.
NVIDIA/Any-to-Any
NVIDIA Releases Efficient Nemotron-3 Multimodal MoE
The new 30-billion parameter Mixture-of-Experts model handles text and images while using only 3 billion active parameters for inference.
Google DeepMind/Any-to-Any
Google Releases Gemma 4 Multimodal Open Model
The new 26-billion-parameter model from DeepMind uses a mixture-of-experts design for greater efficiency and is tuned for assistant-style tasks.
Google DeepMind/Any-to-AnyMajor release
Google Releases Multimodal Gemma 4 31B Model
The new 31-billion-parameter model is an instruction-tuned, 'any-to-any' powerhouse released under a permissive Apache 2.0 license.
Google DeepMind/Any-to-Any
Google Releases 4B Multimodal Gemma 4 Assistant
The new 4-billion-parameter model is instruction-tuned for 'any-to-any' tasks, handling a flexible mix of data types.
Google DeepMind/Any-to-Any
Google Releases 2B Multimodal Gemma 4 Assistant Model
The new compact model from DeepMind is instruction-tuned for "any-to-any" tasks, capable of processing and generating mixed data types.
Xiaomi/Speech → Text
Xiaomi Releases MiMo Model for Speech Recognition
The new open-source model from the Chinese tech giant offers automatic speech recognition for Mandarin, Cantonese, and English under a permissive MIT license.
inclusionAI/Any-to-Any
LLaDA2.0-Uni: A Unified MoE for Vision Tasks
The new open-source model from inclusionAI uses a Mixture-of-Experts architecture to handle multiple vision tasks in a single, diffusion-based system.
DeepSeek/Text / LLMMajor release
DeepSeek Releases V4-Pro, an Open MoE Contender
The new flagship model combines a Mixture-of-Experts architecture with a permissive MIT license, positioning it for wide commercial adoption.
DeepSeek/Text / LLMMajor release
DeepSeek Releases V4-Flash, a Fast MIT-Licensed MoE Model
The new Mixture of Experts model from the Beijing-based AI lab is optimized for fast, efficient conversational AI and carries a fully permissive license.
SenseTime/Any-to-Any
SenseTime Releases 8B Any-to-Any Multimodal Model
The new SenseNova-U1 model unifies image understanding, generation, and editing within a single 8-billion-parameter framework.
Qwen · Alibaba/Vision-Language
Alibaba's Qwen Releases Open 27B Vision Model
The new dense model, licensed under Apache 2.0, brings both text and image understanding to the midrange parameter space.
NVIDIA/Any-to-Any
NVIDIA Releases Nemotron-3-Nano Omni-Modal MoE
The new 30-billion-parameter Mixture-of-Experts model handles any combination of modalities with just 3 billion active parameters.
Resemble AI/Text → Speech
Resemble AI Releases Dramabox Voice Cloning TTS Model
The new text-to-speech model uses a diffusion-transformer architecture for high-quality, expressive audio and one-shot voice cloning.
IBM/Speech → Text
IBM Releases 2B Granite Model for Multilingual Speech
The new two-billion-parameter model offers transcription capabilities for at least five major languages under a permissive Apache 2.0 license.
Qwen · Alibaba/Vision-LanguageMajor release
Qwen Releases 35B Multimodal Mixture-of-Experts Model
The new Qwen3.6-35B-A3B from Alibaba's Qwen team combines vision and language capabilities using an efficient sparse architecture.
Motif Technologies/Text → Video
Motif Releases 2B Open-Source Text-to-Video Model
The new Apache 2.0 licensed model uses a diffusion transformer architecture to offer a new open alternative for video generation research.
Moonshot AI/Vision-LanguageMajor release
Moonshot AI Releases Kimi-K2.6 Multimodal Model
The Chinese AI lab has published weights for its new vision-language model, though a restrictive license limits its use to research applications.
OpenBMB/Vision-Language
OpenBMB Releases MiniCPM-V for On-Device Vision
The new open-source vision-language model is designed for high-resolution image understanding on mobile and edge devices.
MiniMax/Text / LLM
MiniMax Releases M2.7, an MoE Model with FP8 Weights
The new conversational language model from the Chinese AI company uses a Mixture-of-Experts architecture and 8-bit weights, but is released under a restrictive custom license.
Baidu/Vision-Language
Baidu Releases Qianfan-OCR for Document Intelligence
The new vision-language model from the Chinese tech giant is designed for complex, multilingual optical character recognition and layout analysis.
Google DeepMind/Any-to-AnyMajor release
Google Releases Gemma 4, a 26B Vision-Language Model
The new open-source model from DeepMind uses a Mixture-of-Experts architecture to handle both text and image inputs efficiently.
Google DeepMind/Any-to-AnyMajor release
Google Releases Multimodal Gemma 4 31B Model
The new 31-billion-parameter model is instruction-tuned and can process both text and images, marking a significant expansion for the Gemma family.
Black Forest Labs/Text → ImageMajor release
Black Forest Labs Releases 9B FLUX.2 klein Image Model
The new open-weight model offers a more compact, distilled version of the advanced FLUX architecture for text-to-image and editing tasks.
NVIDIA/Vision-Language
NVIDIA's New 3B VLM Pinpoints Objects in Images
The new 3-billion-parameter model, based on the company's Eagle architecture, is designed for high-precision visual grounding tasks.
Google DeepMind/Any-to-AnyMajor release
Google's Gemma 4 Debuts with Any-to-Any Multimodality
The new 4-billion parameter model from Google DeepMind is designed for versatile input and output, handling text, images, and other data types.
Xiaomi/Image Editing
Xiaomi Releases Bilingual Image Editing Model FireRed 1.1
The new open-source model from Xiaomi's FireRedTeam leverages the Qwen-Image-Edit pipeline to offer instruction-based image editing in both English and Chinese.
IBM/Speech → Text
IBM Releases 1B Granite Model for Multilingual Speech
The new Apache 2.0-licensed model is part of the company's Granite family and aims to provide high-quality speech-to-text across several languages.
OpenAI/Text → Speech
Hume AI Releases 3B Multilingual Text-to-Speech Model
The new model, Tada-3B-ML, is designed for fine-grained control over vocal expression across more than 10 languages.
OpenAI/Text → Speech
Kani-TTS-2 Offers New Open-Source Voice Generation
An independent researcher has released a new English text-to-speech model under a permissive license, built on a modern generative foundation.
Zhipu AI/Text / LLMMajor release
Zhipu AI Releases Open-Source GLM-5 MoE Model
The new Mixture-of-Experts model from the Chinese AI company combines an advanced architecture with a fully permissive MIT license for commercial use.
inclusionAI/Any-to-Any
inclusionAI's Ming 2.0 Tackles Any-to-Any Multimodality
The new open-source Mixture-of-Experts model can process and generate content across text, images, and audio in any combination.
Nanbeige/Text / LLM
Nanbeige Releases 3B Chinese-Enhanced Language Model
The new Llama-based model was trained from scratch on 3.5 trillion tokens of Chinese and English data to enhance its bilingual capabilities.
OpenMOSS/Text → Speech
MOSS-TTS: A New Multilingual Text-to-Speech Model
The new system from the OpenMOSS Team uses a novel 'delay-pattern' architecture to generate natural-sounding speech in Chinese, English, and Japanese.
OpenAI/Music
Soul-AILab Releases Zero-Shot Singing Voice Model
The new model, SoulX-Singer, can replicate a singing voice from a short audio sample and supports both English and Chinese under a permissive license.
Zhipu AI/Vision-Language
Zhipu AI Releases Multilingual GLM-OCR Vision Model
The new vision-language model from the creators of the GLM series is specialized for recognizing and extracting text from images across multiple languages.
OpenMOSS/Image → Video
OpenMOSS Releases MOVA for Joint Video and Audio Gen
The new model generates 360p video from text or images and creates corresponding audio tracks simultaneously, a notable step for integrated audiovisual synthesis.
Qwen · Alibaba/Text → Image
Alibaba's Qwen Team Releases Z-Image Diffusion Model
The makers of the popular Qwen language models have published their first open-source text-to-image generator with a permissive Apache 2.0 license.
OpenMOSS/Text → Speech
LuxTTS Delivers Lightweight, Open-Source Speech Synthesis
The new text-to-speech model is optimized for the ONNX runtime, making it a promising option for efficient, on-device audio generation.
Microsoft/Speech → Text
Microsoft Releases VibeVoice for Speech Transcription
The new open-source automatic speech recognition model handles multilingual transcription and speaker identification out of the box.
Qwen · Alibaba/Text → Speech
Qwen Releases Open-Source Voice Cloning Model
The new 600-million-parameter Qwen3-TTS model can generate speech in multiple languages and clone voices from short audio clips.
Qwen · Alibaba/Text → Speech
Qwen Releases a Compact Custom-Voice TTS Model
The new 600-million-parameter model from Alibaba's Qwen team can clone voices from short audio clips for multilingual speech synthesis.
Zhipu AI/Text / LLM
Zhipu AI Releases GLM-4.7-Flash MoE Model
The new Mixture-of-Experts model from the Beijing-based AI company is optimized for speed and released under the permissive MIT license.
Black Forest Labs/Text → Image
Black Forest Labs Releases 9B FLUX.2 Image Model
The new text-to-image model emphasizes speed and efficiency with a novel architecture and FP8 quantization.
OpenMOSS/Text → Speech
Soprano TTS Model Leverages Qwen3 Architecture
The new 80-million-parameter text-to-speech model adapts a powerful language model architecture for efficient, open-source audio generation.
Black Forest Labs/Text → ImageMajor release
Black Forest Labs Releases 9B FLUX.2 Image Model
The new 9-billion-parameter model uses a Diffusion Transformer architecture, promising higher performance than existing open-source alternatives.
OpenAI/Text → Speech
Hume AI Releases TADA 1B for Expressive Speech
The new 1-billion-parameter model combines a Llama 3.2 base with text-to-speech to generate more natural and nuanced audio.
OpenMOSS/Text → Speech
OpenMOSS Releases KugelAudio for European Languages
The new text-to-speech model uses a hybrid diffusion and autoregressive architecture for high-quality, multilingual synthesis.
Zhipu AI/Text → Image
Zhipu AI Releases Open, Bilingual GLM-Image Model
The new text-to-image model is fluent in both Chinese and English, built on the CogView2 architecture and released under a permissive MIT license.
Supertone/Text → Speech
Supertone Open-Sources Supertonic 2 Voice Model
The new text-to-speech model from the audio AI company supports English, Korean, and Spanish and comes in the efficient ONNX format for deployment.
Lightricks/Image → VideoMajor release
Lightricks Releases LTX-2 Multimodal Video Generator
The new diffusion model from the creative app company can generate short video clips from text, images, audio, and even other videos.
Moonshot AI/Vision-LanguageMajor release
Moonshot AI Releases Kimi K2.5 Multimodal Model
The new vision-language model from the Chinese AI firm uses a Mixture-of-Experts architecture and is now available on Hugging Face.
Qwen · Alibaba/Any-to-Any
Qwen's Fun-Audio-Chat: An Open Speech-to-Speech LLM
The 8-billion-parameter model from Alibaba's Qwen team understands and generates spoken responses, enabling more natural audio-first applications.
OpenMOSS/Text → Speech
MiraTTS Brings Qwen2 to Bilingual Speech Synthesis
A new text-to-speech model from OpenMOSS leverages the Qwen2 architecture to generate speech in both English and Chinese.
Qwen · Alibaba/Image Editing
Qwen Releases Open, Bilingual Image Editing Model
The new diffusion model from Alibaba's team allows for precise, instruction-based image modifications in both English and Chinese.
Qwen · Alibaba/Speech → Text
Qwen Releases Compact ASR Model for Streaming Audio
The new Fun-ASR-Nano model from Alibaba's team packs real-time multilingual transcription, speaker diarization, and hotword detection into an efficient package.
Tencent/Image → Video
Tencent's HY-WorldPlay Creates 3D Scenes from One Image
The new model from Tencent's Hunyuan team generates dynamic video and reconstructs 3D environments using a single static picture.
Meituan/Image Editing
Meituan Releases Open, Bilingual Image Editing Model
The new LongCat-Image-Edit model follows natural language instructions to perform complex photo manipulations in both English and Chinese.
Baidu/Image → Video
Baidu's Live-Avatar Animates Photos With Audio
The new 14-billion-parameter model uses audio input to generate realistic talking head videos from a single still image.
Microsoft/Text → Speech
Microsoft Releases VibeVoice for Real-Time AI Speech
The new 500-million-parameter model is designed for generating natural, long-form speech with very low latency for interactive applications.
FlashLabs/Any-to-Any
FlashLabs Releases Chroma-4B, an Any-to-Any Model
The new 4-billion-parameter model handles text, image, and speech inputs and outputs, including direct speech-to-speech translation.
Qwen · Alibaba/Text → Image
Alibaba Releases Z-Image-Turbo, A Fast Open Image Model
The new text-to-image model from the team behind Qwen uses a diffusion transformer to generate high-resolution images in just a single step.
OpenAI/Text → Speech
Nari Labs Releases Dia2-2B, an Open Voice Cloning Model
The 2-billion-parameter text-to-speech model can clone voices from a short audio sample and is available under an Apache 2.0 license.
Allen Institute for AI/Any-to-Any
BAAI Releases Emu3.5, an 'Any-to-Any' Multimodal Model
The new open-source model from the Allen Institute for AI unifies text and image understanding and generation into a single architecture.
OpenMOSS/Text → Speech
SoulX-Podcast 1.7B Offers Open Multi-Speaker TTS
The new 1.7 billion-parameter model from OpenMOSS is trained on conversational data to generate natural dialogue in English and Chinese.
Meituan/Text → Video
Meituan Releases Open-Source LongCat-Video Model
The Chinese tech giant has released a new MIT-licensed model capable of generating video from text, images, or by continuing existing clips.
Meituan/Any-to-Any
Meituan Debuts LongCat-Flash-Omni, an Any-to-Any AI Model
The new open-source Mixture-of-Experts model can process and generate any combination of text, images, video, audio, and 3D data.
NVIDIA/Speech → Text
NVIDIA Releases Real-Time Speaker Diarization Model
The new Sortformer-based model is designed for streaming audio, identifying up to four distinct speakers in real time.
MiniMax/Text / LLMMajor release
MiniMax Releases M2, an Open-Weight MoE for Agents
The Shanghai-based AI startup has released a new Mixture-of-Experts model focused on complex reasoning, coding, and agentic tasks.
Datalab/Vision-Language
Datalab Releases Chandra, a New OCR Vision Model
The new vision-language model from Datalab is fine-tuned from Qwen2-VL to specialize in extracting text and structure from complex documents.
Kuaishou/Any-to-Any
Kling Releases UniVideo for Generation and Understanding
The new open-source model combines both video generation and comprehension, a rare dual capability built on the Qwen2.5 vision-language foundation.
Maya Research/Text → Speech
Maya Research Releases Maya1, an Expressive TTS Model
The new Apache 2.0 licensed model uses a Llama-based architecture to generate more natural and emotionally nuanced speech from text.
DeepSeek/Vision-LanguageMajor release
DeepSeek-OCR Tackles Document Parsing with Vision AI
The new vision-language model uses a novel context compression technique to efficiently extract text and structure from complex documents.
Baidu/Vision-Language
Baidu Releases PaddleOCR-VL for Document AI
The new vision-language model is fine-tuned to understand not just text, but the complex structure of tables, charts, and formulas.
NVIDIA/Speech → Text
NVIDIA's Parakeet ASR Tackles Multi-Speaker Audio
The 600-million-parameter model offers real-time speech-to-text with speaker diarization, built on the efficient FastConformer architecture.
inclusionAI/Any-to-Any
inclusionAI Debuts 'Any-to-Any' Multimodal MoE Model
The new Ming-flash-omni-Preview aims to handle any combination of data modalities using an efficient Mixture of Experts architecture.
Qwen · Alibaba/Vision-Language
Alibaba Releases Qwen3-VL, an 8B Open-Source Vision Model
The latest vision-language model from the popular Qwen series is instruction-tuned and available under an Apache 2.0 license.
Google DeepMind/Text / LLM
Google Releases Compact FunctionGemma Model
The new 270-million-parameter model from Google DeepMind is fine-tuned specifically for reliable function calling and tool use.
EPFL VITA/Image → Video
EPFL Releases SVI for Streaming Image-to-Video
The new open-source model from Swiss researchers uses a novel chunking method to generate indefinitely long videos from a single still image.
Krea/Text → Video
Krea Releases Open-Source Real-Time Video Model
The new 14-billion-parameter model is a distilled, more efficient version of a larger foundation, designed for interactive video generation.
inclusionAI/Any-to-Any
inclusionAI Releases Ming-UniVision MoE Multimodal Model
The new 16-billion-parameter model uses a sparse Mixture-of-Experts design to efficiently handle 'any-to-any' data combinations, from text to images.
Qwen · Alibaba/Vision-LanguageMajor release
Qwen Releases 30B MoE Vision Model, Qwen3-VL
The new open-source model from Alibaba uses a Mixture-of-Experts architecture to make its powerful vision-language capabilities more efficient to run.
nineninesix/Text → Speech
Kani TTS 370M Offers Compact Multilingual Speech
Based on the Language-Free Modeling for Multilingual Text-To-Speech (LFM2) architecture, the new model offers an efficient solution for developers.
chetwinlow1/Image → Video
Ovi Syncs Audio and Video in New Open-Source Model
Built on the Wan2.2 architecture, this new 5-billion-parameter model generates short video clips from a single image and simultaneously creates synchronized audio.
Zhipu AI/Text / LLMMajor release
Zhipu AI Releases Open-Weight MoE Model GLM-4.6
The new Mixture-of-Experts model is available under a permissive MIT license and is optimized for complex reasoning and coding tasks.
inclusionAI/Any-to-Any
Ming-UniAudio Brings MoE to Unified Audio AI
A new 16-billion-parameter model from inclusionAI uses a Mixture-of-Experts architecture to handle a wide range of audio tasks efficiently.
ByteDance/Image → Video
ByteDance Releases Lynx for Identity-Preserving Video
The new model from the TikTok parent company generates short video clips that maintain a person's likeness from a single reference image.
Tencent/Text → ImageMajor release
Tencent Releases HunyuanImage 3.0 Text-to-Image Model
The new text-to-image generator from the Chinese tech giant uses a Mixture-of-Experts architecture for improved efficiency and output quality.
Qwen · Alibaba/Image Editing
Qwen Releases Open-Source Instruction-Based Image Editor
The new model from Alibaba's Qwen team allows users to modify images using natural language prompts instead of complex tools or masks.
Qwen · Alibaba/Any-to-AnyMajor release
Qwen3-Omni Arrives With Any-to-Any Multimodality
The new 30B Mixture-of-Experts model from Alibaba's Qwen team can process and generate content across text, image, and audio formats.
Xiaomi/Any-to-Any
Xiaomi's MiMo-Audio 7B Tackles Complex Speech Tasks
This new instruction-tuned model from Xiaomi can handle a flexible combination of audio and text inputs and outputs, from transcription to voice synthesis.
OpenBMB/Text → Speech
OpenBMB Releases VoxCPM for Open Voice Synthesis
The new 500-million-parameter model offers high-quality text-to-speech and zero-shot voice cloning under a permissive license.
Qwen · Alibaba/Any-to-Any
Qwen Releases 'Thinking' Multimodal MoE Model
The new 30-billion-parameter Mixture-of-Experts model from Alibaba's Qwen team is designed to show its reasoning process for complex multimodal tasks.
Qwen · Alibaba/Any-to-Any
Qwen Releases 30B Model for Audio Captioning
The new Mixture-of-Experts model from Alibaba is fine-tuned to generate detailed, multilingual descriptions for complex audio content.
neuphonic/Text → Speech
Neuphonic Releases NeuTTS Air for On-Device AI Speech
The new Apache 2.0 text-to-speech model is built on a Qwen2 architecture and optimized for local inference with GGUF support.
moondream/Vision-Language
Moondream 3 Arrives in Preview Release
The next generation of the efficient, open-source vision-language model is now available for early testing and feedback.
ByteDance/Image → Video
ByteDance Releases HuMo for Human Video Generation
The new open-source model specializes in creating realistic videos of people, separating appearance from motion for greater control.
Qwen · Alibaba/Text → Video
Alibaba's Wan2.2 Adds Control to Open Video
The new 14-billion-parameter model from Alibaba's PAI team offers fine-grained control over video generation using inputs like sketches and depth maps.
Qwen · Alibaba/Text / LLMMajor release
Qwen Releases 80B Mixture-of-Experts Model
The new Qwen3-Next model from Alibaba combines a large parameter count with an efficient MoE architecture to balance performance and computational cost.
Alpha-VLLM/Any-to-Any
Lumina-DiMOO: A Diffusion Model for Any-to-Any AI
This new open-source model uses a diffusion architecture instead of a typical transformer to generate and understand a mix of media types.
Tencent/Text → Image
Tencent SRPO Fine-Tunes SDXL with Preference Optimization
The new text-to-image model uses a novel rejection sampling technique to align Stable Diffusion XL more closely with human aesthetic preferences.
Tencent/Text → Image
Tencent Releases HunyuanImage 2.1 for Bilingual AI Art
The new text-to-image model from the Chinese tech giant is designed to understand both Chinese and English prompts at high resolutions.
Microsoft/Text → Speech
Microsoft Releases VibeVoice, a 7B Podcast TTS Model
The new 7-billion-parameter model is designed for generating long-form, multi-speaker audio in English and Chinese under a permissive MIT license.
Microsoft/Text → Speech
Microsoft Releases VibeVoice, a Podcast-Ready TTS Model
The new open-source model specializes in generating long-form, multi-speaker audio in both English and Mandarin, mimicking a natural podcast conversation.
StepFun/Any-to-Any
StepFun Releases Step-Audio 2 mini, a Unified Audio AI
The new open-source model handles both speech recognition and audio generation in a single, end-to-end architecture.
Tencent/Image → Video
Tencent's Voyager Model Turns Images into 3D Worlds
The new model from Tencent AI Lab generates temporally and spatially consistent video sequences from a single image, enabling virtual exploration of static scenes.
Microsoft/Text → Speech
Microsoft Releases VibeVoice for Long-Form Audio
The new 1.5-billion-parameter text-to-speech model is designed to generate natural, multi-speaker audio for podcasts and other long-form content.
Qwen · Alibaba/Image → Video
Alibaba Releases 14B Model for Audio-Driven Video
The new Wan2.2-S2V model takes a still image and a speech track to generate a realistic talking-head animation, available under a permissive license.
OpenBMB/Vision-Language
OpenBMB Releases Compact Multimodal Model MiniCPM-V 4.5
The new vision-language model from the open-source research group demonstrates strong OCR and video understanding capabilities in a small package.
DeepSeek/Text / LLMMajor release
DeepSeek Releases 671B MoE Model Under MIT License
The new DeepSeek-V3.1-Base is a massive 671-billion-parameter Mixture-of-Experts model designed for efficient, large-scale research and development.
Qwen · Alibaba/Image EditingMajor release
Qwen Releases Open Model for Image Editing
The new open-source model from Alibaba lets users edit images with simple text commands in both English and Chinese.
NexaAI/Any-to-Any
NexaAI Releases OmniNeural-4B for On-Device AI
The new 4-billion-parameter model is designed for 'any-to-any' multimodal tasks and optimized to run efficiently on mobile hardware.
Tencent/Image → Video
Tencent Releases Controllable Game Video Model
The new Hunyuan-GameCraft 1.0 is an open image-to-video model that generates interactive game-like scenes with precise camera control.
FrancisRing/Image → Video
StableAvatar Brings Open Source Talking Heads to Life
A new diffusion-based model from developer FrancisRing animates still images into talking avatars using only an audio track.
Zhipu AI/Vision-LanguageMajor release
Zhipu AI Releases Open Vision Model GLM-4.5V
The new Mixture-of-Experts model offers strong multimodal reasoning capabilities under a permissive MIT license.
Skywork/Image → Video
Skywork Releases Open 'World Model' for Playable Video
The new 1.3-billion-parameter model functions as an interactive 'world model,' generating controllable video scenes from a single static image.
Google DeepMind/Text / LLM
Google Releases Gemma 3 270M for On-Device AI
The new ultra-compact model from DeepMind is designed for efficient performance in resource-constrained environments like mobile and web.
OpenAI/ReasoningMajor release
OpenAI Releases 21B Open-Weight MoE Model
The new `gpt-oss-20b` is an Apache 2.0-licensed Mixture-of-Experts model designed to run efficiently on consumer-grade hardware.
OpenAI/ReasoningMajor release
OpenAI Releases Its First Open-Source MoE Model
The new 117-billion-parameter `gpt-oss-120b` is a Mixture-of-Experts model focused on reasoning, released under a permissive Apache 2.0 license.
NVIDIA/Speech → Text
NVIDIA Releases Canary 1B v2 Multilingual Speech Model
The new 1-billion-parameter model handles both transcription and translation across five languages using the company's efficient FastConformer architecture.
NVIDIA/Speech → Text
NVIDIA Releases 600M Parakeet for Speech Recognition
The new FastConformer model uses a specialized training technique to improve transcription accuracy in noisy, real-world environments.
Qwen · Alibaba/Text → ImageMajor release
Qwen releases open model for text-in-image generation
The new Apache 2.0 diffusion model from Alibaba's Qwen team focuses on accurately rendering both English and Chinese characters within generated images.
Qwen · Alibaba/Code
Qwen Releases Compact 30B MoE for Coding Agents
The new Apache 2.0 model from Alibaba's Qwen team uses a Mixture-of-Experts architecture to deliver strong performance with only 3B active parameters.
rednote-hilab/Vision-Language
New VLM `dots.ocr` Takes on Complex Documents
The new 3B-parameter model from rednote-hilab uses a vision-language approach to parse tables, layouts, and even mathematical formulas.
Skywork/Any-to-Any
Skywork Releases UniPic, a Unified 1.5B Vision Model
The new autoregressive model from the Chinese AI lab can understand, generate, and edit images within a single, compact framework.
Qwen · Alibaba/Image → VideoMajor release
Alibaba Releases Wan2.2, a 14B MoE Video Model
The new open-source diffusion model from the team behind Qwen uses a Mixture-of-Experts architecture to animate still images.
Qwen · Alibaba/Text → Video
Qwen Releases Wan2.2, a 5B Open-Source Video Model
The new Apache 2.0 licensed model from Alibaba's team generates video from either text prompts or still images, offering a unified approach in a compact package.
Qwen · Alibaba/Text → Video
Qwen Unveils Wan2.2, a 14B Open Text-to-Video Model
The new Apache 2.0-licensed model from Alibaba's team uses a Mixture-of-Experts architecture for efficient, high-quality video generation.
Qwen · Alibaba/Image → Video
Qwen Releases Wan2.2, a 14B Image-to-Video Model
The new 14-billion parameter model from Alibaba's AI team uses a Mixture-of-Experts design and is available under the permissive Apache 2.0 license.
Qwen · Alibaba/CodeMajor release
Qwen Releases 480B Open-Source Model for Code Agents
The new flagship coding model from Alibaba's Qwen team uses a massive Mixture-of-Experts architecture and is released under a permissive Apache-2.0 license.
Zhipu AI/ReasoningMajor release
Z.ai Releases 355B Parameter GLM-4.5 Under MIT License
The new Mixture-of-Experts model combines massive scale with a fully permissive license, targeting complex reasoning and agentic applications.
Qwen · Alibaba/Text → VideoMajor release
Qwen Releases Wan 2.2, a 5B Open Video AI Model
The new Apache 2.0 licensed model from Alibaba's team can generate video from both text and image prompts, adding a powerful new tool to the open-source creative ecosystem.
HiDream.ai/Image Editing
HiDream.ai Releases 17B Open Image Editing Model
The new MIT-licensed model, HiDream-E1.1, allows for complex image modifications by following natural language instructions.
inclusionAI/Any-to-Any
Ming-Lite-Omni 1.5 Brings Any-to-Any Modality to Open Source
The new MIT-licensed model from inclusionAI can process and generate a mix of text, images, audio, and video, pushing the boundaries of open multimodal AI.
RaphaelLiu/Image → Video
Pusa V1: A New Open Model for Image-to-Video Animation
Based on the Wan2.1 architecture, this new 14B parameter model offers fine-grained control over video generation from still images and text.
T-Tech/Speech → Text
T-Tech Releases T-one for Russian Speech Recognition
The new streaming Conformer model from the Russian digital bank is optimized for real-time transcription of telephone conversations.
Moonshot AI/Text / LLMMajor release
Moonshot AI Releases Trillion-Parameter Kimi-K2 Model
The new Mixture-of-Experts model brings massive scale to the open-weights community, focusing on complex reasoning and coding tasks with a 128K context window.
Black Forest Labs/Text → ImageMajor release
Black Forest Labs Releases FLUX.1 Krea Image Model
The new 12-billion-parameter model, tuned by creative AI platform Krea, focuses on high-quality aesthetic output and prompt fidelity.
![FLUX.1 Krea [dev]](/_next/image?url=https%3A%2F%2Fcdn-thumbnails.huggingface.co%2Fsocial-thumbnails%2Fmodels%2Fblack-forest-labs%2FFLUX.1-Krea-dev.png&w=3840&q=75)
ByteDance/Any-to-Any
ByteDance Releases Tar-7B for 'Any-to-Any' Multimodality
The new 7-billion-parameter model from the company's SEED team can process and generate a mix of text, images, audio, and video in a single unified framework.
Boson AI/Text → Speech
Boson AI Releases Higgs Audio v2 for Expressive TTS
The new 3-billion-parameter model focuses on generating expressive, multilingual speech and is fully open for commercial use under an Apache 2.0 license.
Kyutai/Text → Speech
Kyutai Releases 1.6B Bilingual TTS Model
The French AI lab's new open-source model generates streaming audio in English and French under a permissive license.
Zhipu AI/Vision-Language
Zhipu AI Open-Sources 9B Vision Model with 'Thinking' Mode
The new GLM-4.1V-9B-Thinking model makes its vision and chain-of-thought reasoning capabilities available under a permissive MIT license.
AIDC-AI/Any-to-Any
Ovis-U1-3B Unifies Image Understanding and Generation
The new 3-billion-parameter model from AIDC-AI combines vision-language understanding and image generation into a single 'any-to-any' framework.
NVIDIA/Speech → Text
NVIDIA Fuses LLM and ASR in Canary-Qwen 2.5B Model
The 2.5 billion-parameter speech model combines a FastConformer encoder with a Qwen LLM decoder, a hybrid approach to transcription.
Maya Research/Text → Speech
Veena TTS Model Targets Indian Languages with Llama Base
Maya Research has released a 3-billion-parameter model designed to generate natural-sounding speech in Hindi and English.
FreedomIntelligence/Any-to-Any
Janus-4o-7B Adds Image Generation to 7B Multimodal AI
The new 7-billion-parameter model from FreedomIntelligence can process various inputs and generate or edit images based on text prompts.
