Skip to content

AI trending

The 10 most discussed events of the past 48 hours. Each independent source counts once; heat halves every 24 hours.Oct 1 21:35 updated

01−16%6 h

Google DeepMind releases Gemini 4 Argon model

2026年10月1日,Google DeepMind发布前沿模型Gemini 4 Argon。官方发布称,该模型先向Fairwind Program的可信网络防御方开放,后续面向开发者、企业和消费者推出,起步价为每百万输入token 2美元、每百万输出token 10美元。同日The Decoder报道称,该模型在部分基准上超过OpenAI和Anthropic竞品但未取得明显领先,先面向Fairwind计划的“可信网络防御者”和Google内部团队开放,之后才向开发者、企业和消费者推出,尚未公布日期,并提到缓存输入便宜95%。Latent Space报道称,该模型面向编码、企业知识工作与网络防御,支持100万输出token,先在Fairwind Program的政府用户和可信网络防御者中开放。开放对象表述存在差异:官方称先向可信网络防御方开放,Latent Space称先向政府用户和可信网络防御者开放。TechCrunch报道称,Alphabet称其为迄今最强模型,可处理编码、研究和写作等任务,并专门针对防御性网络安全训练,能自主发现、验证和修补关键软件漏洞。

LatestTechCrunch报道称Alphabet称Gemini 4 Argon为迄今最强模型,可自主发现并修补软件漏洞。

Latent Space, TechCrunch · AIand others, 4 sources4 participants

27heat

02−16%6 h

Google pays around 100 publishers for content used by AI

On 30 September 2026, The Decoder and The Verge reported that Google was paying around 100 digital publishers for content used in AI Overviews, AI Mode and the Gemini chatbot, through a pilot launched less than a year earlier. According to The Information, participating publishers are paid according to their content's contribution to these AI features. Revenues vary sharply: several small and medium-sized blogs and websites receive less than 0.1% of their advertising income, with smaller sites earning under $1,000 over several months. Some publishers have received $50,000–$60,000 over a few months, while one early participant earned more than $1 million in a year. On 1 October, Ars Technica reported that several smaller sites put these payments at roughly one-thousandth of their advertising revenue, and some large sites were refusing to participate while seeking better terms. Google had long resisted direct payments to websites, arguing that traffic in exchange for scraping content already constituted an exchange.

Ars Technica · AI, The Verge · AIand others, 3 sources3 participants

15heat

03−15%6 h

OpenAI DevDay 2026 announces Dots and GPT-6.1 Sol

On 29 September 2026, OpenAI held DevDay 2026 at Fort Mason in San Francisco. Ahead of the event, The Verge reported that OpenAI was trailing in consumer AI agents and expected it to unveil its own agent, Aeon, competing with Meta’s Muse, SpaceX’s Grok Bot and OpenClaw. OpenAI later published an official recap covering more than 20 updates across GPT-6 Astra, ChatGPT, Codex, APIs and safety, but provided only an overview. Simon Willison liveblogged the keynote. From 30 September, TechCrunch and The Decoder reported that ChatGPT plugins had expanded into app-like experiences with a sidebar entry, interactive panels and file viewers, and that Codex had introduced cloud development environments reusable across devices. ChatGPT also gained the Plugin Extensions system, shared Space, collaborative Pages and Slides, Slack and Microsoft Teams integrations, automated workflows and an enterprise marketplace. TechCrunch said colleagues in Space could collaborate with chatbots and the new agent role Dots. OpenAI announced more than 1.2 billion weekly active ChatGPT users, over 35 million combined weekly active users for ChatGPT Work and Codex, and 2.5 million enterprise customers. Latent Space reported the launches of persistent Dots agents, GPT-6.1 Sol, Ultrafast mode, the Decisions API, ChatGPT Spaces and Marketplace. Earlier reports expected Aeon, whereas later reports described Dots; the initial official recap mentioned GPT-6 Astra, while later reports named GPT-6.1 Sol.

Latent Space, TechCrunch · AIand others, 4 sources4 participants

13heat

Ranks 4 to 10

  1. 04

    Trump and tech CEOs sign voluntary AI pledge

    On 30 September 2026, Trump signed an AI code of conduct at the White House with tech executives including Meta CEO Mark Zuckerberg, OpenAI’s Greg Brockman, Nvidia’s Jensen Huang and Elon Musk. According to The Decoder, it carries only a moral obligation. That evening, The Verge reported that the full agreement, formally named the Joint Commitment on Frontier Responsibility, had been published and shared online by David Sacks. Signatories included Google’s Sundar Pichai. Earlier reports listed Musk among the signatories, while later reports named Google’s Pichai, Anthropic’s Dario Amodei, Meta’s Zuckerberg and OpenAI’s Brockman. Reports have not yet detailed the code’s provisions or enforcement mechanisms, and the story is still developing.

    The Verge · AI, The Decoder2 sources2 participants

    10heat

    −16%6 h
  2. 05

    Altman says OpenAI won't go public before model safety commitments

    OpenAI CEO Sam Altman previously said the company is unlikely to go public this year. On 30 September 2026, in a press Q&A after the DevDay keynote, Altman went further, saying the company will not go public before it can make reliable commitments on model safety, and that there is no clear timeline. He said waiting too long would also be bad for the world, and that he worries going public would create pressure to disappoint Wall Street supporters in the name of safety. Altman explained "pacing the frontier" as putting safety and alignment ahead of capability, rather than directly slowing down. On 1 October 2026, TechCrunch reported that at the same Dev Day, Altman announced the launch of the Decisions API, which lets users preset a set of options for the Luna model to classify or choose agent behaviour, released as a limited preview.

    TechCrunch · AI, The Verge · AI2 sources2 participants

    9heat

    −15%6 h
  3. 06

    OpenAI DevDay unveils GPT-6.1 Sol and always-on agent Dots

    On 29 September 2026, OpenAI unveiled two products at its DevDay keynote in San Francisco. Dots is an always-on AI assistant that performs tasks across connected apps in the background and gradually learns users' preferences. It has its own cloud computer, can access a browser and over 4,000 supported apps, and is not tied to specific hardware or interfaces. GPT-6.1 Sol reportedly offers near-Astra intelligence in agentic coding, computer use and professional work, with API input and output token prices one-fifth of Astra's standard rates. Both initial and subsequent reports said GPT-6 Astra powered Dots. Initial reports said Astra's release had been delayed over safety issues, while later reports described it as the flagship GPT-6.1 Astra. Dots initially opens only to paying ChatGPT Pro, Business Premium and Enterprise subscribers, competing with Meta's recently launched Muse.

    TechCrunch · AI, The Verge · AIand others, 3 sources3 participants

    8heat

    −16%6 h
  4. 07

    OpenAI discusses a $30 billion funding round at a $1.4 trillion valuation

    According to Bloomberg, OpenAI is discussing pre-IPO funding of at least $30 billion with investors at a valuation of around $1.4 trillion. The report said a strategic focus on key areas such as coding had driven run-rate revenue up 70% since July, reaching $40 billion in August. Ars Technica later reported that OpenAI had delayed its IPO plans over AI safety concerns and was instead seeking another $30 billion in private funding. Talks remain ongoing, with no investor list, specific terms or completion date disclosed.

    Ars Technica · AI, TechCrunch · AI2 sources2 participants

    8heat

    −16%6 h
  5. 08

    Anthropic says Zhipu GLM-5.3 exploit capabilities approach Claude

    On 30 September 2026, Anthropic published an assessment finding that Zhipu's open-weight GLM-5.3 model approached Claude Mythos Preview in exploit development capabilities. That day, the Anthropic Frontier Red Team released details: across 100 random tasks in its internal Binary Exploitation benchmark, GLM-5.3 achieved full control-flow hijacking in 4% of trials, versus 6% for Claude Mythos Preview. Earlier models such as Claude Opus 4.6 and GLM-5.2 had not succeeded on these tasks, and the report said a meaningful capability threshold had been crossed. Initial reports mentioned the close capabilities without methods, scores or comparisons; subsequent reporting supplied the 4% and 6% figures and benchmark details. No response from Zhipu has been reported.

    The Decoder, Simon Willison2 sources2 participants

    8heat

    −16%6 h
  6. 09

    AMD acquires World Labs for $8.2 billion

    On 29 September 2026, several outlets reported AMD's announcement that it would acquire Fei-Fei Li's World Labs for $8.2 billion. TechCrunch said the deal was expected to close by year-end, subject to regulatory approval, with Li joining AMD as executive vice-president and chief scientist. The Verge described an approximately $8.2 billion all-stock deal, noting that World Labs was founded in 2024, reached a $1 billion valuation within months and launched its first commercial product, Marble, in 2025 to generate interactive 3D worlds from prompts. Latent Space added that since its founding, World Labs had assembled model training teams for images, video and spatial reconstruction, and advanced robot simulation through its acquisition of SceniX. Its recently released Atlas is an omni model architecture that predicts novel views from 2D images, combining generative models and multi-view geometry to address the long-standing sparse reconstruction problem in computer vision. On 30 September, The Decoder reported that Li would report directly to CEO Lisa Su and lead frontier research, while Ars Technica described the move as a stronger challenge to Nvidia. No other details of the deal's structure have been disclosed.

    Ars Technica · AI, The Decoder2 sources2 participants

    6heat

    −16%6 h
  7. 10

    GPT 6.1 Sol launches at one-fifth of Astra's price

    On 29 September 2026, Simon Willison commented that GPT 6.1 Sol offered intelligence close to Astra at one-fifth of the price. The article was tagged OpenAI, generative AI, LLM and GPT, and was recorded as the latest development at 02:27 on 30 September. At 03:34 that day, the AWS Machine Learning Blog reported that GPT-6.1 Sol was generally available on Amazon Bedrock, with stronger reasoning for agentic coding, computer use and everyday professional workloads. According to OpenAI, it matches GPT-6 Astra on DeepSWE v1.1 at around one-fifth of the cost per task and exceeds GPT-6 Sol's best result by 6.4 percentage points. The confirmed facts are its release and Bedrock availability, a price around one-fifth of Astra's, and assessments of near-Astra intelligence. Reports provided no further details on specific parameters, launch time or official pricing.

    AWS Machine Learning Blog, Simon Willison2 sources2 participants

    6heat

    −17%6 h
How heat is counted

Heat comes from the independent accounts and organisations taking part in one event. Repeated collection counts once, and heat decays with a 24-hour half-life. It measures how active the discussion is, not how good the reporting is.

The ranking covers the past 48 hours. The trend compares only the same set of sources we have followed throughout, so it reflects our coverage, not the whole web. Without comparable history, no trend line is shown.

The list of sources shows only reports that can be read in public; participants also include accounts and organisations counted only for heat. Several channels of one organisation may be counted together, so there are not always more participants than sources. Open an event to read each side's reports and views.

Surging
Discussion is growing fast
New
First reported within 6 hours
Rising
Discussion is still growing