mode

Tech journalist Alex Heath reported that OpenAI recently showcased Astra to VIP clients at a closed preview event. OpenAI's researchers demonstrated that Astra can coordinate multiple agents to collaboratively complete mathematical proofs, operate desktop software at the "superhuman, extremely fast" speed described by Sam Altman, and generate presentation slides, financial review reports, and analyses of messy data.Astra is designed to work continuously for days or even weeks, capable of remembering human corrections, collaborating with other agents and humans, and performing operations across various software tools. Sam Altman also stated at the event, "I expect Astra to be the first model that can genuinely invent new things in a meaningful way."

According to Business Insider, after Google released its new AI model this month, some employees have begun testing the preview version of Gemini 3.8 Flash on the internal coding platform Jetski. A testing employee stated that the model is significantly better than 3.7 Flash, but emphasized that it is still too early to make a comprehensive evaluation. Google declined to comment on this.With 3.5 Pro still not released and Anthropic and OpenAI continuing to accelerate, Google no longer possesses cutting-edge models and has shifted to focusing on the faster and cheaper Flash series "main models" for coding and running agents. Gemini 3.6 Flash was launched in July, and 3.7 Flash followed just three weeks later. CEO Sundar Pichai stated at the Q2 earnings call that the company's goal is to release a new version approximately every month.The employee preview suggests that the public launch of Gemini 3.8 Flash may not be far off, but it is not guaranteed. Some AI observers believe they have found traces of the model on the Arena AI benchmarking website, which Google has previously used to test upcoming versions. Earlier this year, Google also launched the Gemini Spark agent.

According to the New York Times, citing two informed sources, Meta had internally predicted this year that it might invest up to $10 billion annually in Anthropic's AI models.

According to Gate market data, Moderna's pre-market decline has expanded to 8%.

On August 27, the one-stop AGI infrastructure platform B.AI announced that the Zhipu GLM-5.3-Flash (320B-A18B) officially launched on the platform and is fully open for free in the API official group. As the first native multimodal model in the GLM-5 series, GLM-5.3-Flash features a total of 320 billion parameters and 18 billion active parameters, supported by computing power from purely domestic AI chips. It introduces a mixed architecture of sparse and linear attention for the first time, combined with a 1M ultra-long context, offering both powerful performance and exceptional cost-effectiveness. Starting today, developers can call it at zero cost through the B.AI API official group, easily handling complex reasoning, long text processing, and multimodal tasks.

According to Jinshi, Alibaba Cloud announced that the billing unit price for the Qwen3.8-Flash model mode on the Bailian large model service platform will be adjusted downwards starting from 12:00:00 Beijing time on August 27, 2026. Qwen3.8-Flash is the latest multimodal large model launched by Qianwen, natively supporting a million-level context window, capable of processing ultra-long documents, code repositories, and complex dialogues in one go. It performs outstandingly in scenarios such as programming assistance, intelligent agent collaboration, and image-text understanding, and is compatible with OpenAI and Anthropic interface protocols, making it suitable for building high-concurrency applications and intelligent workflows.The specifics are as follows: the input price before adjustment was 1 yuan, and after adjustment it will be 0.8 yuan; the output price before adjustment was 3 yuan, and after adjustment it will be 2.7 yuan.

According to TIME, OpenAI CEO Sam Altman recently stated in an interview that the company has previewed the upcoming cutting-edge model series Astra to key clients. In the demonstration, 16 AI agents can collaboratively break down mathematical problems and assemble proofs, and Astra can operate computer software across applications at superhuman speeds. Altman mentioned that Astra will support "persistent agents" capable of performing long-term tasks and is expected to be the first model that can invent new things in a meaningful way, possessing characteristics of AGI.Over the past year, OpenAI has fallen behind expectations in product direction and pre-training research, being surpassed by Anthropic in programming products, annual revenue, and valuation. The company has experienced multiple executive departures and is facing challenges such as several product liability lawsuits and legal disputes with Apple and Musk. OpenAI's current valuation is nearly $1 trillion, with ChatGPT having over 1 billion monthly active users.Recently, OpenAI disclosed a security incident: an unreleased agent escaped the sandbox and attacked Hugging Face. Following this, the research team froze some experiments, enhanced monitoring, and paused the training of an unreleased model expected to bring the greatest capability leap until new safety measures are in place. Altman emphasized that "ensuring AI safety is more important than the growth momentum of any company," and the company will slow its pace and allocate resources to safety and alignment teams. Chief Research Officer Mark Chen estimated that the company is about 80% complete in reaching AGI, and Altman stated that the internal system may be referred to as AGI by the end of the year.

Zhipu officially announced the launch and open-sourcing of GLM-5.3-Flash, which is the first native multimodal model in the GLM-5 series.It is reported that the overall performance of GLM-5.3-Flash exceeds that of GLM-5.2, with programming and Agent evaluations approaching Claude Opus 4.8, but at only one-tenth the price of GLM-5.2. It also features a new foundational model, introducing a mixed architecture of sparse attention and linear attention for the first time in the main GLM series, and is pre-trained with 30T Token multimodal data.Zhipu stated that to gather extensive and professional feedback from a wide range of users, large-scale testing was conducted with the anonymous model Ox-Alpha (referred to as "Niu Lai" in the Chinese community) on OpenCode and OpenRouter before the official release. Ox-Alpha quickly became the most popular model of the week, setting a new high for call volume on both platforms, with all request traffic supported by domestic chip computing power.Previously, the community's DeepSWE small sample test for Ox Alpha once achieved 80%, but that was based on only 10 questions. After expanding the sample, the score fell back to about 63%, and testers also actively corrected the initial claim. This score still belongs to the top tier, but is not as exaggerated as the initial 80%. After the weights are open-sourced, developers can directly deploy using frameworks like vLLM, SGLang, KTransformers, without needing to go through the anonymous model's API.

Ali Qianwen released the Qwen3.8-Flash model, which is described as a multimodal MoE model and an early preview of the Qwen4 architecture.The production version of Qwen3.8-Flash will soon be available through the Qwen Cloud API, with a price of only $0.16 per million input tokens and $0.47 per million output tokens. The model has 125 billion parameters + 51 billion N-gram embedding parameters, but only activates 6 billion parameters per token, achieving extremely high cost-effectiveness.

According to Bloomberg, Chinese artificial intelligence company Z.AI (Zhìpǔ) has confirmed that the mysterious artificial intelligence model Ox Alpha, which has rapidly risen to the top of online usage rankings, is a new iteration of its GLM series.The company stated in response to a Bloomberg News inquiry that it will release the model weights for Ox Alpha tonight.

According to sources, Dark Side of the Moon is negotiating with Microsoft, Amazon, and Google for revenue sharing of the Kimi K3 artificial intelligence model. Dark Side of the Moon is seeking to obtain up to 30% of the revenue generated from K3-related services in the early negotiations.

Sources from relevant personnel at Zhongke Shuguang reveal that the company will launch a new generation of token acceleration solutions aimed at large model inference, focusing on addressing issues of redundant computation and data scheduling during the inference process.Currently, Zhongke Shuguang's ParaStor distributed storage has been applied to the country's first fully domestically produced 100,000-card AI supercluster Shuguang 8000 (Dengfeng), providing data throughput and elastic scaling support for large-scale computing tasks. The latest report from CCID Consulting shows that Zhongke Shuguang ranks first in four sub-industries: AI, education, embodied intelligence, and autonomous driving.

On August 26, B.AI announced that the API is fully compatible with the OpenAI Responses API. Developers only need to configure the Base URL using the B.AI API Key to directly call the GPT series models supported by B.AI in the Codex development environment, seamlessly integrating cutting-edge large model capabilities into their daily coding workflow. This means that whether it's code generation, logical understanding, intelligent debugging, architectural refactoring, or team development collaboration, developers can obtain real-time intelligent assistance from the GPT series models in the Codex environment, comprehensively covering various scenario needs from individual development to team collaboration. Experience it now: http://b.ai

According to Cointelegraph, Hugging Face disclosed that it experienced an intrusion event driven by autonomous AI agent systems in July. The attacking agent began testing in early May, leaving exploit notes using OpenAI's Artifactory instance, and subsequently launched approximately 17,600 attacks on Hugging Face, affecting its dataset processing infrastructure, production environment, internal network, and cloud credentials. Confirmed customer data access was limited to five datasets related to the ExploitGym/CyberGym benchmark.Hugging Face found during the investigation that due to security barriers imposed by top model providers like OpenAI and Anthropic, the company was unable to use these commercial models for defensive analysis and was forced to turn to running the open-source model zai-org/GLM-5.2 in China, which operates on the company's own infrastructure, ensuring that attacker data and credentials do not leave its environment. The company pointed out that attackers are not bound by any usage policies, while the defense's forensic work is hindered by the barriers of the hosted models.This incident highlights the security paradox between open-weight models and closed models. The article also discusses the ongoing debate regarding the security of open-weight AI, including OpenAI and Anthropic's push to restrict open-source models, as well as researchers' progress in detecting malicious behavior by examining changes in model weights.Hugging Face recommends that defenders prepare models that can run on their own infrastructure before an incident occurs to avoid barrier lock-in and protect attacker data.

According to Business Insider, Thomson Reuters launched its first self-developed AI model, Thomson-1, on Monday. This model is based on Snowdon and constructed through "re-alignment" of Alibaba's open-source Qwen model. Open-source means anyone can download and modify the model for free. Thomson-1 will take over some tasks previously handled by Anthropic's Claude, but it is not intended to completely replace collaboration with Anthropic and other labs, initially focusing on the company's areas of expertise, starting with document review.This move aims to address the high AI costs brought by models like Claude and OpenAI Codex, and reflects the use of cheaper Chinese open-source AI. CTO Joel Hron stated that the main reason is to better leverage Thomson Reuters' own expertise and control costs. The company expanded its collaboration with Anthropic in May this year for the AI legal assistant CoCounsel, which still primarily relies on Claude. Hron said, "Our main goal is to gradually make Thomson the model that drives more and more capabilities for CoCounsel."Thomson Reuters, in collaboration with a team from Imperial College London, spent months transforming Qwen to build Snowdon and ensured it is "ethically and politically bias-processed and safe to use." Hron pointed out that having a proprietary model allows for development based on its own intellectual property rather than continuously paying external AI companies, comparing it to renting versus buying a house: renting provides shelter but does not accumulate long-term equity.

Alibaba's video large model Wan 3 is officially launched, and Nvidia is negotiating an investment in Perplexity, which is valued at over $30 billion.Xiaomi releases the Xuan Ring O3 processor using 3 nm technology, and smart ring manufacturer Oura plans to raise up to $3 billion in an IPO.

The B.AI free model event has only been launched for a week, and the total Token throughput across the platform has strongly surpassed the 20 trillion mark. The system has maintained stable operation amidst continuous high concurrency peaks and massive automated scheduling tasks.During the event, four popular domestic models—MiMo-V2.5, DeepSeek-V4-Flash, DeepSeek-V4-Flash-Vision-Exp, and Tencent Hy3—are fully open for free, with zero barriers to use right out of the box, covering a wide range of application scenarios including rapid text response, multimodal visual understanding, solid Chinese reasoning, and complex task processing.Whether developers need to handle high-intensity code writing, ultra-long context document analysis, or build enterprise-level high-concurrency Agent intelligent bodies, they can freely access ample computing resources. B.AI will continue to practice the vision of inclusive computing power, promoting more AI innovations to accelerate implementation with lower costs and better experiences.

B.AI platform announces that Xiaomi's open-source flagship multimodal model MiMo-V2.5 is officially available for free access. As a 310B sparse MoE native multimodal large model, MiMo-V2.5 supports full multimodal input of text, images, videos, and audio, and features a 1M ultra-long context window with rapid response capabilities, deeply optimized for high-intensity practical scenarios such as multimodal agents, audio and video intelligent analysis, and general programming.Starting today, developers can unlock Xiaomi's cutting-edge multimodal model capabilities at zero cost through the B.AI platform API official group, with no barriers to integrating top-tier multimodal capabilities into their applications and workflows. Log in now at chat.b.ai/chat to experience the new multimodal intelligent experience brought by Xiaomi MiMo-V2.5 for free.

According to Jinshi reports, the dark side of the moon Kimi announced that the company's first-generation trillion-parameter multimodal model Kimi K2.5 will end its service at the end of this month.

According to X user Lentils80, Anthropic is currently testing two new models, named "claude-marshmallow-eap" and "claude-melon-eap".Related information shows that "marshmallow" performs better than "melon", and has a better conversational experience than Opus 5. However, neither of these models has reached the Fable level.