Large Model

On November 14, New Wave Weibo released its first large open source model, VibeThinker-1.5B, called “Small models can also be intelligent”. 1AI is accompanied by the following official presentation: Are the most powerful models in the industry at present more than 1T, or even 2T-scale models, that have a high level of intelligence? Is only a small number of technology giants capable of making large models? VibeThinker-1.5B, which is exactly the negative answer given by Weibo AI to this question, proves that small models can also have high IQs. It's..

On November 4th, the company LongCat Magna Carta announced that the official App has now been officially released to support online search and can also launch voice calls (video calls are expected). 1AI query learned that LongCat App is now available online in Android and iOS versions, which can be downloaded through LongCat official network, and apple users can search for "LongCat" directly in App Store. "LongCat 1.0" on line! Free experience with global lead, AI model..

On October 24, news that yesterday, in the Alpha Arena challenge of the AI Large Model De facto Exchange, Ali Baba, under the banner of Qwen3-Max, won the championship for the first time with efficient tactics and precision, breaking DeepSeek’s long lead. This competition, initiated by the American Artificial Intelligence Research Laboratory, Nof1.ai, provides real money of $10,000 each to the six top American models for real-time investments on the Hyperliquid Exchange. The participation model is Claude 4.5 S..

On 25 September, according to the news, at the 2025 Cloud Habitat Congress, the Director of the Ali Group and CEO Wu Yui Song indicated that the 38 billion AI infrastructure was being actively advanced and that additional inputs were planned. At the conference, Ali launched six new models + 1 brands: Qwen MAX: Large Model for trillion Parameters, Coding and Tool Call Capability to Top the International List; Qwen3-Omni: Large Model for a New Generation of Numerical Whole Models, which truly achieves the "Qwen3-VL:Agent and Coding"

According to news from September 9, the Chinese Academy of Sciences Automation Institute's Official Micro-List Public News, recently the Chinese Academy of Sciences' Automation Institute, Li Kuoqi, and Team Seo Bo, working in collaboration with MétaX, developed the Great Brain Pulse Model “SpikingBrain-1.0”, completed process training and reasoning on the national production kilocalories GPU computing platform, achieved quantitative efficiency and speed enhancement of the Big Model in super-long sequence reasoning, and demonstrated a new type of non-Trans that built national production autonomy and controls..

On September 4, Shanghai Artificial Intelligence Laboratory (Shanghai AI Lab) announced the open source of general multimodal large model Shusheng Wanxiang 3.5 (InternVL3.5), which upgraded its reasoning ability, deployment efficiency and generalization ability. InternVL3.5 has nine sizes of models with parameters ranging from 1 billion to 241 billion, which can meet the needs of various scenarios. Among them, the flagship model InternVL3.5-241B-A28B scored 77.7 points in the multidisciplinary reasoning benchmark MMMU, which is the most ...

August 16 news, according to Xinhua News Agency reported today, the Shenzhou 20 astronaut crew on the night of 15 successfully completed the third out of the cabin activities, China Astronaut Research and Training Center, said that in the preparation of out of the cabin activities, a large model called "Wukong AI" played an auxiliary support role for the astronauts in orbit to provide intelligent, specialized support. It provides intelligent and specialized support for the astronauts' work in orbit. A month ago, on July 15, the Tianzhou-9 cargo spacecraft was successfully launched, sending the astronauts a new generation of space suits, fresh peaches, and a large model called "Wukong AI" based on the domestic open-source model. The large model has been in orbit for a month...

July 29 news, according to the central video reported today, in the 2025 China AI grand ceremony site, a man-machine debate intense staged, and around the debate of the "Turing test" also successfully passed. Chen Ming, a debater from Qibao, debated with MiniMax's big model debater, and more than 30% viewers were "confused" by the AI, and according to the rules of the "Turing Test", the test was passed. Note: The Turing test is a thought experiment proposed by British computer scientist Alan Turing in 1950, the procedure of which is for an inquirer to write down...

July 27 news, today in the 2025 World Artificial Intelligence Conference during the Shangtang big model forum, Shangtang Technology officially released Rizhixin V6.5 big model and "Wuneng" body intelligence platform. It is reported that Rizhixin V6.5 big model has been upgraded to a greater extent in the reasoning aspect, and with the improvement of multimodal performance, the cost has dropped significantly. According to 1AI, "Wuneng" Embodied Intelligence Platform is based on Shangtang's embodied world model as the core engine, relying on Shangtang's large devices to provide end-side and cloud-side computing power support, and is able to provide robots and intelligent devices with powerful perception, visual navigation and multimodal...

July 27 news, according to CCTV news reports, this year China's basic big model iteration speed up, big model in electronics, raw materials, consumer goods and other industries to speed up the landing. The reporter learned from the World Conference on Artificial Intelligence that the current total number of released big models in the world has reached 3755, of which, China's enterprises have contributed 1509 (Note: the proportion of 40%), the number of which is the first in the world. According to 1AI's previous report, China Internet Network Information Center (CNNIC) released the 56th Statistical Report on China's Internet Development in Beijing in July this year. The report...

June 29 news, according to CCTV news reports, the reporter learned from the National Energy Group, China's first professional power generation industry large model -- "Prime Source" released in Beijing, the model parameters of the hundreds of billions of levels, which is also the world's first large model of hundreds of billions of levels of power generation industry, effectively improving the reasoning ability of the model to achieve safe, efficient, green and intelligent power generation to provide a "super brain". This is also the world's first large model of the power generation industry with hundreds of billions of parameters, which effectively improves the reasoning ability of the model and provides a "super brain" for the power generation industry to realize safe, efficient, green and intelligent power generation. 1AI learned from the report that the "Prime Source" large model is based on the volume of power installed capacity of the National Energy Group, which is the world's largest, relying on a powerful unified AI platform, integrating high-quality data sets unique to the power generation industry, and overall planning...

June 6, 2011 - Beijing Zhiyuan Artificial Intelligence Research Institute (BJAII) released a series of "Wujie" models, including the world's first native multimodal world model "Wujie・Emu3", the world's first multimodal generalized basic model of brain science Emu3", the world's first native multimodal world model, and "Brainμ", the world's first multimodal universal basic model for brain science. WisdomSource also released Wujie/Embodied Intelligence models, including the world's first MCP-enabled cross-body brain-size collaboration framework, Wujie/RoboOS 2.0, the embodied brain model, Wujie/RoboBrain 2.0, and the all-atom microscopic life model, Wujie/OpenComplex2, which are the first to support MCP.

On May 28th, the Ant-Ling Big Model (Ling) team officially open-sourced the unified multimodal big model Ming-lite-omni. According to the introduction, Ming-lite-omni is an all-modal model based on the MoE architecture constructed by Ling-lite, with a total parameter of 22B and an activation parameter of 3B, which supports "cross-modal fusion and unification", "understanding and unification", and "generation unification". It supports "cross-modal fusion and unification" and "comprehension and generation unification". In several comprehension and generation evaluations, Ming-lite-omni with only 3B of parameter activation, performs as well as 10B of...

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...

May 10, 2011 - IBM on May 2 introduced a preview of one of the smallest versions of its Granite 4.0 family of models: Granite 4.0 Tiny. The Granite 4.0 Tiny Preview benefits from high computational efficiency and low memory requirements: at FP8 precision, running a concurrent dialog of five 128KB context windows requires only 12GB of graphics memory, and a single NVIDIA G...

May 8, 2011 - Mistral launched Medium 3, a mid-weight version of its Mistral 3 series of models, on May 7, local time, touting both cutting-edge performance and significantly lower cost with a wide range of enterprise features. Mistral claims that Medium 3 meets or exceeds Anthropic's Claude Sonnet 3.7's 90% performance in every benchmark, and costs only 0.4 per million Token output ...

April 30 news, Xiaomi big model team announced through the "Xiaomi MiMo" public number, today, Xiaomi open source the first "born for reasoning" big model Xiaomi MiMo, linkage pre-training to post-training, comprehensively improve the reasoning of the The first model of Xiaomi MiMo, linking pre-training to post-training to improve reasoning in all aspects. According to the introduction, MiMo is the initial attempt from the newly established "Xiaomi Big Model Core Team". In the open evaluation sets of mathematical reasoning (AIME 24-25) and code competitions (LiveCodeBench v5), MiMo outperforms ...

April 10 news, "SenseNova V6" multimodal fusion big model released today, with the longest 64K chain of thought, mathematical analysis, multimodal depth inference, global memory and other capabilities, support for 10-minute video inference and depth inference. According to 1AI, SenseNova is a large model system announced by SenseNova, which includes a natural language processing model "Consultation" (SenseChat), a text-to-diagram model "Second Painting" and a digital human video generation platform "Ruying". SenseAvatar, a platform for generating videos of digital people...

March 30, according to Xunfei medical news, the national "four major chronic diseases" major special core results -- the world's first "type 1 diabetes disease model" was officially released yesterday. It is said that this is a leap from the laboratory to the clinic, but also in Anhui Province in the field of chronic disease prevention and treatment for the first time to realize the results of national major scientific research projects. According to reports, the project focuses on type 1 diabetes diagnosis and treatment pain points, through the integration of multimodal data and clinical experience, based on the Xunfei Starfire medical model X1, constructed to cover the screening, typing, treatment and management of the whole chain of digital intelligence prevention and control system. 1AI from KU Xunfei official...

March 17 news, @ millet technology official microblogging said today, millet big model team in the field of audio reasoning breakthrough progress. Inspired by DeepSeek-R1, the team took the lead in applying reinforcement learning algorithms to multimodal audio comprehension tasks, and in just one week it topped the internationally authoritative MMAU audio comprehension evaluation list with a SOTA accuracy of 64.5%, which is now open sourced at the same time. 1AI's official article reads: Reinforcement Learning Shows "Counterintuitive" Advantage -- Xiaomi's Big Model Team Tops MMAU Audio Reasoning Chart Faced with a video of a car driving...

March 11, OpenAI today announced its latest research, trying to use CoT (chain of thought) monitoring way to monitor the reasoning model's "ideas", so as to stop the AI big model nonsense, hide the real intention and other behaviors, at the same time, this is also one of the effective tools to supervise the super model. Note: AI will occasionally talk nonsense, which we generally call AI hallucinations. In addition to this, AI can be lazy or deliberately bad, such as tricking the user or simply giving up when a problem is too difficult. OpenAI uses the newly released cutting-edge model o3-mini as the monitored...

February 24 news, the current global technology giants power AI, e-commerce platform companies have also joined. According to Lei Feng network news today, Jinduoduo internal has formed a large model team of e-commerce recommendation, the person in charge of the core members of Baidu Phoenix Nest before. According to sources, this person was previously in Baidu when it is already ten million annual salary, and this time Pinduoduo give the sincerity is also very sufficient, its annual package is several times in Baidu. Another source said that Pinduoduo recruited a group of people from Baidu last year to do big model related projects. At present, Pinduoduo big model products are divided into several applications: price comparison system, recommendation, advertising, search...

Feb. 24 - Clement Delangue, co-founder and CEO of Hugging Face, a community for AI big models, announced in a post early this morning that DeepSeek-R1 has received more than 10,000 likes on Hugging Face, making it the most popular big model out of nearly 1.5 million models on the platform. "It's astounding!" Late last month, Hugging Face also announced the Open-R1 project, which plans to fill the DeepSeek R1 un...

Since the Spring Festival, DeepSeek's heat continues to climb, accompanied by a lot of misunderstandings and controversies, some people say that it is "the light of the national product to beat OpenAI", and some people say that it "is just a copy of the foreign model of the work of the small smart". These misunderstandings and controversies are mainly focused on five aspects: 1, excessive myths and brainless disparagement, DeepSeek in the end is not the underlying innovation? Is there any basis for the so-called distillation of ChatGPT? 2, the cost of DeepSeek, really only 5.5 million dollars? ...