Kunlun Wanwei

On August 11th, Kunlun World Wide officially released SkyReels-A3 model. Officially, SkyReels-A3 model is based on "DiT (Diffusion Transformer) video diffusion model + frame insertion model for video extension + action optimization based on reinforcement learning + controllable mirror operation", which can realize full-modal audio-driven digital human creation of any duration. It is reported that SkyReels-A3 model brings new experience to users in the following four directions: Text Prompt (Text Prompt Input) supports screen...

August 12, 2011 - Kunlun World Wide's SkyWork AI Technology Release Week was launched on August 11, with one model released every day for five consecutive days, covering the core multi-modal AI scenarios. Today, Kunlun brings the upgraded version of Matrix-Game interactive world model in Matrix series of self-developed world model - "Matrix-Game 2.0", which is claimed to be a world model that realizes interactive real-time long sequence generation in general scenarios. In order to promote the development of interactive world modeling, "Matrix-Game 2.0" is said to realize the interactive real-time long sequence generation world model in general scenarios.

August 3 news, Kunlun Wanwei announced the launch and open source of a new reasoning big model Skywork MindLink. According to the introduction, this new reasoning paradigm, able to adaptive integration of reasoning and non-reasoning according to the difficulty of the task to generate a response, greatly reducing the cost of reasoning, and enhance the readability and effectiveness of the reasoning process. Data show that Skywork MindLink model in 10 in no external tool configuration of the large model review comparison, in the human last exam (Humanity's Last Exam) review won...

July 9, 2011 - Kunlun World Wide has just released an announcement announcing the launch of the latest Skywork-R1V 3.0 version and open source. According to KLM, Skywork-R1V 3.0 deeply stimulates the cross-modal reasoning ability of the model through reinforcement learning strategies in the post-training phase, achieving a double leap in complex logic modeling and cross-disciplinary generalization. Skywork-R1V 3.0 is based on the previous generation of inference model, Skywork-R1V 2.0, which distills the data for "cold start", and builds high-quality multimodal inference training by rejecting sampling...

May 26, 2011 - Kunlun World Wide Group today announced the official launch of the Kunlun World Wide Skywork Super Agents (Skywork Super Agents) App. This is the world's first Office Intelligent Body mobile app based on AI Agent architecture, marking the era of "AI Office Intelligent Body" from the desktop to the mobile terminal. According to the introduction, Skywork Super Agents App is the world's first mobile intelligent body platform covering full-scene office and multimodal creation, supporting three expert-level intelligent bodies and a general intelligent body...

May 22, 2011 - Kunlun World Wide today launched Skywork Super Agents for the global market. The product adopts AI agent architecture and deep research technology, and is capable of one-stop generation of documents, PPTs, tables (excel), web pages, podcasts, and audio/video multimodal content. According to the report, its deep research capability is ranked No. 1 in the world on the GAIA list, surpassing OpenAI Deep Research and...

May 13, 2011 - Kunlun World Wide has just announced the official open source of Matrix-Game big model (17B+), the interactive video generation big model in Matrix-Zero world model. Kunlun says that Matrix-Game is the official landing of Matrix series in the direction of interactive world generation, and the first open source 10B+ spatial intelligence big model in the industry, which is an interactive world base model for game world modeling, designed for high-quality generation and precise control in open environments. According to the introduction, M...

April 21, 2011 - Kunlun World Wide Web today officially released and open-sourced SkyReels-V2, the world's first unlimited duration movie generation model using the Diffusion-forcing framework, which achieves synergistic optimization by combining Multimodal Large Language Modeling (MLLM), Multi-stage Pretraining, Reinforcement Learning, and Diffusion-forcing frameworks. It achieves co-optimization by combining MLLM, Multi-stage Pretraining, Reinforcement Learning, and Diffusion-forcing frameworks. Officially...

On April 13th, Kunlun World Wide announced that the Skywork team has launched the newly upgraded Skywork-OR1 (Open Reasoner 1) series of models based on the iterative optimization of Skywork-o1, the first large Chinese logical reasoning model. According to the introduction, this series of models has realized industry-leading reasoning performance under the same parameter scale, further breaking through the bottleneck of large models in logic understanding and complex task solving. At the same time, Skywork-OR1 will be fully open and free to use, and give back to the developer community in the form of full open source. The ...

On March 26, 1AI learned from Kunlun World Wide that Mureka O1 and Mureka V6 models were officially released today, the former is the world's first music inference big model, officially said its performance surpassed Suno, topped SOTA. Mureka V6 is the current Mureka's base model, supports pure music generation, and also supports AI music creation in 10 languages, including English, Chinese, Japanese, Korean, French, Spanish, Portuguese, German, Italian and Russian. In Mureka V6, the group...

February 18, 2011 - Kunlun World Wide announced today that it has open-sourced SkyReels-V1, the first video generation model for AI short drama creation in China, and SkyReels-A1, the first SOTA-level video base model-based expression action control algorithm in China. Kunlun World Wide's official introduction said that SkyReels-V1 marks the details of the performances, and also processes the emotions, scenes, and performance demands. SkyReels-V1 marks performance details, processes emotions, scenes, performance demands, etc., and utilizes "ten million-level, high-quality" Hollywood-level data for training and fine-tuning. In addition, SkyReels-V1 can realize "movie and TV level characters...".

On November 27th, Kunlun World Wide Group announced the official launch of Skywork o1, a series of models with complex thinking and reasoning ability. According to the introduction, Skywork o1 is a series of models with "slow thinking and reasoning ability" released by Kunlun World Wide Group, and it is the first o1 model with Chinese logical reasoning ability in China. Different from the existing work of reproducing OpenAI o1 models, Skywork o1 not only endogenizes the ability to think, plan, reflect, etc. in the model output, but also in the standard evaluation ...

Nov. 20, 2011 - This morning, Kunlun World Wide announced the launch of "Skywork 4.0" version 4o (Skywork 4o) and a new product "real-time voice dialogue assistant Skyo". According to the official introduction, Skyo, as an intelligent voice interactive product, has the ability of fast response and multi-language dialog, it can initiate dialog, interrupt in real time, overcome the "illusion" problem of the big model, and reply to the real content in the dialog. At the same time, Skyo has emotional response and personalized voice customization functions. "Skyo version 4o has the following features...

Kunlun World Wide officially announced today that its series of models with complex thinking and reasoning ability -- Skywork O1 version 4.0 (English name: Skywork O1) will start inviting tests on November 27th. According to the official introduction of Kunlun World Wide, Skywork O1 version 4.0 is the first o1 model in China with Chinese logical reasoning ability, which supports thinking, planning, reflecting and other abilities in the model output. Tiangong Big Model version 4.0 O1 includes two models: the open source model based on the open source Llama 3.1 8B, which...

Kunlun Wanwei announced that its AI short drama platform SkyReels will be launched in the US on December 10th. According to the official introduction, in terms of script generation, SkyReels has enriched the creative templates of explosive models. Among them, the script big model generation ability has been greatly improved, especially optimized the short drama's cool point generation ability, and the current generation effect can reach A or even S rating in the artificial script rating on average. In terms of character generation, the R&D team has increased the AI actor library. At the same time, the R&D team has constructed actor attribute labels, and based on the multimodal big model capability, it can intelligently help users find the most suitable...

Kunlun Wanwei Technology Co., Ltd. recently launched the music platform "Yinfeng AI". This service can not only create music of various styles based on the text submitted by users, but also generate complete songs up to 4 minutes long to meet users' needs for music length. The "Yinfeng AI" platform provides users with a brand new music creation experience. Users only need to provide a piece of text, and AI can convert it into melody and rhythm to create personalized music works. In addition, the platform also provides partial generation and extension functions, and users can rewrite or extend a certain melody to obtain more satisfactory music effects. …

According to Kunlun Wanwei, SkyReels, the world's first AI short drama platform that integrates large video models and 3D models, was officially released today. According to the official introduction, the SkyReels platform integrates script generation, character customization, storyboards, plots, dialogues/BGM and film synthesis. Creators can "make a drama with one click". Users input a simple idea, and the tool can complete the entire process from script to finished short drama production. The official shared a demo of an AI short drama as follows: One day, the sensitive and introverted boy Xiaojie was walking alone on the street, but he accidentally met a...

Kunlun Wanwei announced on August 14th, Beijing time, that it has officially launched the world's first AI streaming music platform Melodio, and simultaneously launched the AI music commercial creation platform Murek. According to reports, users can input prompts in Melodio according to the current scene or mood, and Melodio will continue to generate customized music of the corresponding style. During the process of Melodio music generation and playback, users can modify the input copy at any time, change the generated music content, and can also query the generated lyrics in real time, and select the favorite clips...

In order to help developers efficiently switch to domestic large models, Kunlun Wanwei Tiangong Open Platform (https://model-platform.tiangong.cn/) immediately launched the "One-click Migration Plan" for OpenAl API users, and provided 150 million Tokens of API call resources (including Tiangong 2.0 and Tiangong 3.0 large models)! Developers will also enjoy exclusive Tiangong migration service technical support. Tiangong Open Platform will provide migrating users with: 150 million Tokens of API call Tiangong 3.0A...

Against the backdrop of the rapid development of large model technology, Kunlun Wanwei has open-sourced a landmark sparse large language model, Skywork-MoE. This model not only performs well in performance, but also significantly reduces the cost of inference, providing an effective solution to the challenges brought by large-scale intensive LLMs. Skywork-MoE model features: Open source and free for commercial use: Skywork-MoE's model weights and technical reports are completely open source and free for commercial use without application. Reduced inference cost: While maintaining strong performance, the model significantly reduces...

Kunlun Wanwei announced today that it has open-sourced the 200-billion-dollar sparse model Skywork-MoE. This is based on the intermediate checkpoint of the Skywork-13B model that Kunlun Wanwei previously open-sourced. It is claimed to be the first open-source 100-billion-dollar MoE model that fully applies and implements the MoE Upcycling technology. It is also the first open-source 100-billion-dollar MoE model that supports reasoning with a single RTX 4090 server (8 RTX 4090 graphics cards). According to reports, the open-source Skywork-MoE …

Kunlun Wanwei announced that the daily active users (DAU) of its Tiangong AI has exceeded one million. This achievement not only reflects the enthusiasm of the majority of users for Tiangong AI, but also indicates that Kunlun Wanwei's technology and product capabilities in the field of AI have been widely recognized by the market. Since the launch of the Tiangong AI big model, Kunlun Wanwei has always insisted on user experience as the center, continuously optimizing product experience, and creating AI products that are closer to user scenarios. Based on the continuous breakthroughs in model technology, Tiangong AI has provided users with AI search, AI music, AI painting, AI writing and other dimensions...

On the evening of April 23, Kunlun Wanwei disclosed its 2023 annual performance report. The company achieved operating income of 4.92 billion yuan in 2023, of which overseas business income accounted for 86.0%, an increase of 8.4 percentage points year-on-year; net profit was 1.26 billion yuan, an increase of 9.15% year-on-year, continuing to maintain the industry's leading profit scale. The company firmly implements the strategy of "all in AGI and AIGC" and increases R&D investment. In 2023, the company's R&D expenses increased by 40.2% to 970 million yuan. The company…

Kunlun Wanwei announced today that the performance of the Tiangong 3.0 big model has been significantly improved, and its Tiangong SkyMusic music big model has also been open to the public for public testing today. Tiangong 3.0 has 400 billion parameters, surpassing Grok-1 with 314 billion parameters, and is the world's largest open source MoE big model. Tiangong 3.0 has significantly improved performance in semantic understanding, logical reasoning, versatility, generalization, uncertainty knowledge, learning ability, etc., and its mathematical/reasoning/code/cultural and creative capabilities have been improved by more than 30%. Tiangong 3.0 …