Open Source

On 29 July, according to TechCrunch, Anthropic CEO Dario Amodei published a book on local time on 27 July, responding to outside rumours that Anthropic supported a ban on open-source weight models. In Bo, Amodei made it clear that "Anthropic has never advocated the prohibition of open-source weight models" and stressed that open-source weight models that do not have a dangerous capability are "a public good" and can bring value to businesses, developers and researchers. He also pointed out that even if he came..

On July 28th, the topic of whether AI should open up weight has recently become the focus of industry discussions. Last Friday, more than 20 companies, including Ying Weida, Meta and Microsoft, signed an open letter entitled Open Weights and American AI Leadership in support of Open Weights. By the end of the week, OpenAI, Google and SpaceX also signed the letter. Anthropi..

News from July 26, Andrew Ng (24 July) released a dynamic in LinkedIn, announcing the launch of OpenWorker, an open-source AI smart body, not only for chatting, but also for the efficient performance of the client's task. On the running platform, OpenWorker for Apple MacOS is on line, and the Github page shows Windows versions supported but not yet signed. OpenWorker based on the original..

On July 23rd, Inweida CEO Wong In-hoon, in an interview with Axios, stated that the United States had no fear of China’s open source AI models, and should be more vigilant in demanding that they be banned. He called the Chinese models "excellent" and that good open source models should be used, as should United States enterprises. This statement follows the release of Kimi K3 on the dark side of the month, and the renewed discussion of China’s open model risks in American politics. OpenAI and Anthropic accused Chinese competitors of distilling their model capabilities and alerting Washington to..

On July 8th, the Hong Kong Big Data Intelligence Laboratory (HKUDS) recently released the Open Source AI Smart Body Mentor DeepTutur v. 1.5, which completed several upgrades on document decomposition engines, RAG intake pipe lines and Python 3.14+ compatibility. DeepTutor's core design is to integrate the six components of chat, testing, research, visualization, problem resolution and learning path planning into the same smart body closed loop, so students do not need to interrupt backstage engines when they switch between assignments, and learning context and multilevel memories automatically coincide..

On July 2nd, the Portuguese government announced on July 1st that the first open-language model, AMALIA, based on Portuguese in Europe, was officially launched. 1AI understands that, as a national strategic project in Portugal, the AMALIA model was developed over a period of 18 months by more than 60 researchers from various academic institutions in Portugal, with an advance investment of 5.5 million euros, using such computing infrastructure as Deucalion and Mare Nostrum 5. The first stage of its development was the training of about 4 trillion Portuguese words..

On June 25th, the Internet announced yesterday the launch of the TTS Voice Synthetic Engine Confucius4-TTS. Officially, this is the first open-source model in the industry to support 14 languages with no accent and without reference to text. Confucius 4-TTS supports voice cloning of zero samples. 3 seconds of audio material provided by users, without reference to text and prior training, the model completes acoustic cloning; officially known as the cloned sound is more than 85% and the cloning mission is as accurate as..

On June 23rd, yesterday, Kyoto announced an open source live video visual language interactive model, JoyAI-VL-Interaction. Officially, this is the first all-inputing model and system in the world and is supported by vLLM-Omni from day-0. JoyAI-VL-Interaction supports voice input output, visual interface, long-term memory, back-office model interface and vLLM deployment programme. Kyoto claims that developers can replace ASR, TTS, backstage models, external..

On June 19, according to news, yesterday, Chairman Cai Chongsing of the Ali Baba Group, at the European Science and Technology Summit VivaTech 2026, described the development strategy of Ali Baba All-Ai. Cai Zhongsin argued in a conversation that the total potential market size of AI (TAM) should target human productivity itself. “AI is creating value units that are equivalent to human intelligence and human productivity. And globally, at least half of the 100 trillion dollars of GDP -- 50 trillion dollars, about human productivity and human intelligence, that's the total of AI..

On June 18, according to the news, the think-tank announced yesterday to be online and open-source GM-5.2, the new model focused on Coding and the long-range mission. GLM-5.2 has an automated research, performance optimization, complex system engineering, and deep debugging Coding Agent scenario training; In the FrontierSWE long-range mission assessment, GLM-5.2 Max 10 Hrs score 72.6%, lower than Claude Opus 4.8 Max 20 Hrs..

On June 16, MiniMax announced yesterday that the model weight of MiniMax M3 has been officially launched under the open-source flag, and that the MSA (MiniMax Sparse Act) technical paper has been published simultaneously. Mini Max M3 Total Parameter 428B, Activating Parameter 23B, officially positioned as the first open-source model to introduce multi-modular hybrid training from Step 0. The training phase was heavily integrated with text, images and multi-modular staggered data for pre-training levels..

On June 16th, a Rio 3.5 Open 397B model under the umbrella of the municipality of Rio de Janeiro, Brazil, has recently been the subject of discussion in the AI community. The model page shows that it is open in the MIT protocol, with a total parameter of about 397B, single token activates about 17B and supports a longer context. However, the model was published less than 24 hours ago, and the open-source AI team, Nex-AGI, spoke on platform X, pointing out that Rio 3.5 was actually consolidated through the model (Model Mer..

ON MAY 23RD, THE INTERNET ANNOUNCED YESTERDAY THAT IT WAS DECIDED TO FORMALLY TARGET THE GLOBAL FULL-SCALE OPEN SOURCE FOR THE "SUB" LARGE MODEL, THE "MULTI-MODULAR MODEL" AND THE "SPOKEN SYNTHESIS (TTS) MODEL, THE TWO-ENGINE CORE ENGINE OF 4.0. DEVELOPERS CAN DOWNLOAD, DEPLOY AND RE-DEVELOP ON THIS BASIS FREE OF CHARGE. THIS OPEN-SOURCE "SUB" MULTI-MODULAR MODEL (27B PARAMETER SIZE) IS ORIENTED TOWARDS THE EDUCATIONAL SCENE, SUPPORTING THE MATHEMATICAL CAPABILITY OF VISUAL INPUT AND ACHIEVING THE TOP INDUSTRY LEVEL (SOTA). IN THE SIZE MODEL OF THE SAME PARAMETERS, THE DIFFICULTY OF VISUAL MATHEMATICAL PROBLEMS WITH A CHART IS ADDRESSED. CHINESE NUMBER OF PLAIN TEXT..

Message of May 14, yesterday, the face wall intelligence officially released and opened with a source of 1.3B end-side polymodular megamodel MiniCPM-V 4.6, providing two versions of Instract and Thinking. Instruct version: missions for common image understanding, STEM mathematical reasoning, OCR documentation, video time-series understanding and target positioning, going beyond Qwen3.5-0.8B and Gemma4-E2B-it; Thinking version: high-level tasks for multi-image association reasoning, hallucinating inhibition..

News of 8 May, yesterday, Mi AI laboratory released and opened-source multi-language voice cloning TTS model OmniVoice, and the team built 580,000 hours of training data based on 50 open-source data sets covering 646 languages. (b) In 24 languages, speech similarities and understandings go beyond multiple commercial systems

On April 27th, according to blogger @foodshop researcher Will, on today's Mi Investor Day, the Vice-President of the Mi Group and Chairman of the Technical Committee Tsumi gave a speech on the theme “Agent-type remodeling of rice and human-carnage ecology”. The organizers summarized the core view as follows: Agent remodels the paradigms of millet and `human-carrying all-ecosystems' and spends more than $60 billion on AI over the next three years, a figure that is just below the bottom and higher; Mi has the full capacity of the IA age: the base, the data, the model, the frame and the ecological layer; the remodeling of one side..

Message from April 24, this morning, the DeepSeek-V4 model preview was officially online and synchronized with the open source. DeepSeek-V4 has a million-word super-long context, leading both domestic and open source areas in Agent capabilities, world knowledge and reasoning. The model is divided into two versions by size: the entry of the official network chat.deepseek.com or the official App will allow dialogue with the latest DeepSeek-V4 to explore the full new experience of 1M memory of the extra-long context. API Service Synchronized More..

On April 23rd, the Ali Yuntung team announced yesterday that the open source model family had received new members — Qwen 3.6-27B. This is a dense, multi-modular model with 27 billion parameters and the highest community call for model specifications. This was followed by the release of Qwen3.6-Plus and Qwen3.6-35B-A3B, and this open-source version of 27B, while maintaining the advantages of a dense architecture, has resulted in a comprehensive upgrading of intelligent body programming and multimodular reasoning. According to official sources, Qwen 3.6-27B supports..

In April 21st, yesterday, the dark side of the moon officially released and opened a new model, Kimi K2.6, which focused on upgrading the code, AI intelligence and office capacity. (a) Stronger code capability: the internal benchmark test score is about 201 TP3T, which can be coded continuously for 13 hours and handles over 4000 line codes to support multiple languages such as Rust, Go, Python

Hermes Agent is an open source AI Agent framework published by Nous Research, authorized by the MIT protocol. Unlike IDE's programming assistant or a single API chat robot packager, Hermes Agent is an autonomous intelligence deployed on a user-owned server with a permanent memory of a session and automatic skill generation capability - the framework automatically extracts the solution path as a reusable skill file, calls and continuously optimizes the next time you encounter a similar task, forming &qua..

Step 1: What is Hermes Agent? Hermes Agent is an open-source AI Agent developed by the Nous Research team. It's not like a regular chat robot (such as ChatGPT), it's going to: remember all your previous conversations (long-term memory) and learn from each mission, automatically create new "skills" to help you do real things: execute orders, browse web pages, write files, manage tasks, etc., support connections to Telegram, flying books, Discord, etc...

IN THE NEWS OF APRIL 9, YESTERDAY, ISP AI OFFICIALLY ANNOUNCED THE OPEN SOURCE FLAG-LEVEL AI SMART BODY ENGINEERING MODEL GLM-5.1 AND CONTINUES TO FOLLOW A LIBERAL MIT LICENSE TO SUPPORT PERSONAL AND COMMERCIAL USES. GLM-5.1 IS THE STRONGEST FLAGSHIP MODEL PUBLISHED SO FAR IN THE BRAIN SPECTRUM, WITH THE CORE DESIGN OBJECTIVE OF SUSTAINED AND EFFECTIVE IMPLEMENTATION OF AI SMART BODY TASKS OVER A LONGER PERIOD OF TIME. IT WAS DESCRIBED THAT GM-5.1 WAS ABLE TO WORK INDEPENDENTLY AND CONTINUOUSLY FOR MORE THAN EIGHT HOURS IN A SINGLE MISSION, DURING WHICH IT WAS AUTONOMOUS IN PLANNING, IMPLEMENTING, AND PLACING ITSELF IN AN ITERATIVE MANNER, LEADING TO THE DELIVERY OF COMPLETE ENGINEERING-LEVEL RESULTS. IN OFFICIAL..

In the morning of April 3, Google DeepMind released a new generation of open source model series Gemma 4 with a one-time rollout of four models covering the side to the workstation scene. E2B: 5.1 billion total parameters, 2.3 billion valid parameters, 128K context, officially stated as some equipment memory occupancy up to below 1.5GB; E4B: 8 billion total parameters, 4.5 billion valid parameters, 128K context, MMLLU Pro, 69.4%, close to previous generation 27B level; 26B A4B M..

Jellyfish is a one-stop AI-generated comic video production tool that supports the automatic generation of spectroscopy, role, scene and full video from novel text. The overall location is the "AI Shorts Factory". Its core objective is to upgrade short-time production from a manual/semi-automatic mode to an industrial stream. Not only is the tool fully open-source, supporting local deployment and secondary development, but it also strikes at the technical level more accurately at the worst point of pain for the generation of AI video. The Jellyfish feature Enteres the script: Just provide text scripts in support of Chinese and English. Smart Mirror..