NTrending

AI trending

Trending

Last 48 hours, most discussed AI stories, top 20 stories

Oct 30 21:50 Updated
NO.01

Google DeepMind releases the Gemini 4 Carbon model

On October 28, 2026, Google DeepMind released Gemini 4 Carbon, calling it its strongest model, ranking first with 59 points on the Artificial Analysis Intelligence Index v5.0, and scoring 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench. The model is aimed at long-horizon engineering tasks; internal teams once had it run continuously for several days to investigate and fix hidden defects in a large codebase, and it rose to 4th place on WebDev Arena with 1768 points. Carbon was first opened to some cybersecurity defenders, then gradually expanded through the Fairwind Program, and afterward made available to developers, enterprises, and consumers; after the introductory period, pricing is $4 per 1 million input tokens and $20 per 1 million output tokens. On October 29, Arena.ai said Carbon ranked 2nd in Agent Arena based on real tasks, with a net improvement of 16.2%, and was on the Pareto frontier for cost, time, and output tokens per task. On the same day, Yuchen Jin cited Artificial Analysis data to say Carbon is smarter than Opus 5.5 and costs less per task, but noted that the evaluation used Antigravity

NO.02

Mistral releases Large 4 weights on Hugging Face

On October 29, 2026, Mistral AI released the open weights of Mistral Large 4 (Le Chonk), a 1-trillion-parameter native multimodal model that continues reinforcement learning on the basis of the preview version, with improved coding and defensive cybersecurity capabilities. According to data published by Mistral AI, the model scores 19.7% on FrontierCode 2.0 and 48% on the Artificial Analysis Coding Agent Index, roughly on par with GLM-5.3, but uses an average of 47% fewer tokens per task; on Terminal-Bench 4.0 it improves by 13 percentage points to 41%. The model supports three reasoning modes, uses a modified MIT Mistral Ecosystem Open-Weight license, and is priced at $0.70 per million input tokens, $0.07 for cached input, and $2.10 for output via the API. On the same day, Unsloth released a 1-bit GGUF quantized version of the model, enabling it to run locally on a Mac Studio, and said the quantized version still retains 91% top-1 accuracy.

16 sources · 18 participants
103
NO.03

Anthropic Publicly Submits S-1 Registration Statement for Proposed IPO on Nasdaq

On October 26, 2026, Anthropic publicly submitted its Form S-1 registration statement, planning to list on Nasdaq under the ticker ANT. The prospectus shows revenue of $16.286 billion in the first half of 2026, up nearly 13-fold year-over-year, with Q2 operating profit of $974 million for the quarter. Subsequently, on October 27, World Trade Securities estimated that its cash reserves could approach $400 billion by the end of 2027; Anthropic reiterated in the cover letter of its IPO filing that Claude will not display ads, and on October 28 acknowledged that its life sciences business lacks physical laboratory experience. Goldman Sachs secured the lead-left bookrunner seat on the prospectus, while Morgan Stanley moved to the role of stabilizing market agent. TechCrunch reported that after the IPO, public shareholders will have almost no say over board candidates. On October 29, the prospectus disclosed four previously unreported Agent safety incidents. On October 30, Steve Eisman accused the prospectus of deliberately withholding third-quarter data, and Anthropic has not yet responded; Reuters reported the same day that its IPO fundraising cap could reach $100 billion.

55 sources · 15 participants
92

More stories No.04–20

Participants · 24-hour trend · Heat index

  1. 04

    OpenAI doubles the usage limits for Pro 100/200/500 and Business Premium

    On October 30, 2026, OpenAI announced that effective immediately, usage limits for all Pro 100/200/500 tiers and Business Premium will be doubled. The company stated that after listening to feedback, it will no longer reduce usage for Pro 200, usage for Work and Codex will remain unchanged, the extended quota for dots will continue to be offered for the time being, and dot usage will still be calculated separately from the default usage. On the same day, Vaibhav Srivastav further clarified that users who have already subscribed to Pro 200 will retain their existing Work and Codex usage quotas and will not see them reduced. He had previously said that the usage limits for Pro 100/200/500 and Business Premium would double starting that day, and that the Pro 200 quota would not be lowered.

    8 sources79
  2. 05

    Zhiyuan releases GO-3 Preview, AHA3 and G2 Max

    On October 29, 2026, Zhiyuan released its new-generation embodied intelligence foundation model GO-3 Preview at Innovation Day, along with the AHA3 hardware platform and the industrial robot G2 Max. According to a report by Shangguan News, GO-3 Preview connects the entire process from training and simulation practice to continuous learning after going on duty, ranking first in three tests—RoboDojo, RoboTwin, and RoboColiseum—and is planned to be fully open-sourced by the end of 2026; in a real-world convenience store scenario test, the Elf G2 robot's task success rate was 28.6% on the first day and rose to 85.7% five days later. The next day, Zhiyuan co-founder Yao Maoqing said in an interview with Securities Times that general large models have obvious shortcomings in physical robot operation, and that future systems will move toward a layered architecture; he also mentioned that GO-3 Preview adopts the WRAM native architecture, raising the zero-shot average success rate on 100 tasks across 20 skill categories from 13.4% to 56.4%.

    8 sources62
  3. 06

    Alphabet's third-quarter revenue and earnings per share both fell short of expectations

    Alphabet reported third-quarter adjusted earnings per share of $2.67 and revenue of $127.58 billion, both below analysts' expectations. According to the Google Blog, revenue for the quarter grew 25% year over year, with annualized revenue reaching $500 billion, and Google Cloud revenue grew 103%; CEO Sundar Pichai said the first-party model API processes nearly 32 billion tokens per minute, 40% more than the previous quarter, and the Gemini app has reached 1 billion users. Solidot reported that the stock price fell nearly 10% after the earnings release, and the company disclosed that revenue generated by AI search is lower than that of traditional search, with overall search business revenue growing 12% year over year to $63.25 billion; Chief Business Officer Philipp Schindler said the slowdown in search business growth stems from the lower monetization capability of AI Mode. Spending for the quarter was $51.4 billion, and net profit was $26.17 billion, lower than $35 billion in the same period last year.

    12 sources45
  4. 07

    Baidu upgrades search AI API, minimum 4.9 yuan per thousand calls

    At its product upgrade communication meeting on October 30, Baidu announced a fully upgraded search AI API, with a minimum price of 4.9 yuan per thousand calls during the promotional period, lower than the price of the DeepSeek search API. According to The Beijing News, Baidu said the capability now covers more than 170 tool scenarios, with GUI scenario coverage on the search page increased fivefold, access to over 5 billion resources of data including government affairs, livelihood, and the central bank, coverage of more than 20 industries, and an open ecosystem; its cost-effectiveness comes from an AI-native four-layer architecture, where the engine layer self-evolves through multi-Agent collaboration, raising overall iteration efficiency by more than 80%, and the application layer supports full-solution access via API, MCP, and Skill. TMTPost·AGI reported the same day that the API's content layer covers more than 80 languages and a high-quality index in the hundreds of billions, with data credibility improved by over 50%, and that the Ernie Assistant saw a 20% performance improvement after integration. There are currently no further developments.

    3 sources44
  5. 08

    Tsinghua University, together with Infinigence AI and AgiBot, proposes a collective experience learning paradigm

    On October 30, 2026, AgiBot, together with Tsinghua University and Infinigence AI, released the Collective Experience Learning (CEL) paradigm, which enables robots to use their peers' real task trajectories as learning material, with humans only responsible for judging the outcomes. The paradigm is accompanied by the FERT failure-guided experience retrieval algorithm and the FATS fault-tolerant asynchronous training system. On five real-robot tasks—flower arrangement, clothes folding, phone charging, stool assembly, and phone packaging—the policies trained with CEL all exceeded the average level of six human teleoperators, leading by up to 16.7 percentage points, with an overall success rate of 97.5%. In the phone packaging task, the number of robots participating in training expanded from 16 to 96, and the success rate rose from 86.7% to 98.3%. In the 96-robot experiment, FATS completed a cumulative 9,834 rounds of task collection, and the machine-time loss caused by 30 failures accounted for only 1.3% of the cluster's total operating time.

    6 sources41
  6. 09

    Anthropic trains Claude to flag potential abuse in conversations

    Anthropic has trained Claude to perform a safety self-assessment of the entire conversation after each response, in order to flag potential abuse in the conversation. According to Rohan Paul, this technique, called self-assessment, has since its launch in June detected abusive behavior that other safeguards failed to cover; the mechanism leverages the advantage that Claude has already read the conversation, and costs less than a classifier. Phoenix Technology, citing Anthropic's safety blog, reported that models released from Claude Fable 5 onward will append an instruction invisible to users after the response ends, reusing existing context and the KV cache and adding less than 2% in compute overhead. On Fable 5, the mechanism captured 78% of violating conversations that a heavy-duty classifier could flag, with a 94% detection rate for high-risk abuse and a 1% false-positive rate; the assessment results are not shown to users and do not enter subsequent conversations, and are mainly used for offline aggregate screening, while real-time blocking remains the responsibility of other safety classifiers.

    4 sources39
  7. 10

    Cognition Launches Coding Benchmark FrontierCode 2.0

    On October 28, 2026, Cognition launched the coding benchmark FrontierCode 2.0 to evaluate whether the code produced by models can actually be merged; even the highest-scoring Opus 5.5 scored only 45%, showing the benchmark is far from saturated. According to the Cognition blog, compared with version 1.1, the reference solutions modify on average 67% more files and 89% more lines of code, the codebase size is 25% larger, task coverage repositories increased by 55%, and blocking scoring criteria increased by 43%; DeepSeek V4.1 Flash dropped from 43.9% to 19.3%, and Kimi K3 dropped from 44.2% to 17.7%. On the same day, Artificial Analysis released Intelligence Index v5, listing FrontierCode 2.0 as one of the newly added benchmarks, with 60% of evaluations based on private test sets or held-out answers; Claude Opus 5.5 (max) topped the list with 58 points.

    7 sources37
  9. 12

    Gemini app launches in-chat file generation and editing

    On October 30, 2026, the Gemini app launched the ability to generate, preview, and edit files within chats. According to the Google Blog, users can type edits directly in full-screen mode within a conversation, select text to ask questions about it, or have Gemini rewrite the file; PDF preview and editing are supported on both web and mobile, while other formats such as Google Docs, Google Sheets, Slides, Markdown, TXT, and code are supported only on web. The feature is available to all users in all languages covered by the Gemini app.

    7 sources34
  10. 13

    Binance hot wallet stolen in EIP-7702 malicious authorization attack

    In the early hours of October 30, 2026, a hot wallet labeled "Binance 14" by Binance was submitted 50 EIP-7702 authorizations; 3 minutes and 24 seconds later, 167,464 ETH, 1.615 billion USDT, and 63 types of tokens were swept away, with cross-chain losses of $5.5 billion to $6.2 billion. The market then swung sharply, with Bitcoin falling below $78,500 and Ethereum losing $2,250, and $629 million in liquidations across the network in one hour. Issuers including Tether and Circle froze the related assets, and Binance verified that the total frozen amount was $3 billion. Shenyu's review stated that no contract vulnerability was seen on-chain, the private key is still in Binance's hands but the malicious authorization has not been revoked, and the hacker may have induced the signing of a Type-4 transaction in the signing or operations pipeline. The incident is still being handled.

    1 sources34
  11. 14

    Aaron Levie says the gap between the AI frontier and the general public continues to widen

    On October 30, 2026, Box CEO Aaron Levie offered his assessment, arguing that the gap between AI frontier practitioners and everyone else is not only failing to narrow but is continuing to widen, because the AI capability curve is rising almost vertically. He pointed out that processes solidified 3 to 6 months ago may already be outdated today, that companies that completed their transformation a year ago face an even larger gap, and that AI-native companies therefore hold a clear advantage, with the pressure falling on whether large companies can keep up. On the same day, Rohan Paul, citing Levie's view, commented that the gap between those currently working at the AI frontier and everyone else is almost impossible to comprehend, and that this gap curve is not converging but instead continuously expanding. At present, this assessment was put forward by Levie and relayed by Rohan Paul, with no further developments yet.

    2 sources33
  12. 15

    Apple launches touch-enabled OLED MacBook Pro

    On October 29, 2026, Apple launched the touch-enabled OLED MacBook Pro, its first touchscreen Mac. According to Computerworld, the 14-inch version starts at $2,899 and the 16-inch version starts at $3,399, with sales beginning on November 6; the new machines are equipped with M5 Pro and M5 Max chips, up to 128GB of unified memory and 16TB of storage, and Apple says the M5 Max can run large language models with hundreds of billions of parameters entirely on-device, with touch functionality integrated into macOS 27. 9to5Mac reported the same day that the new machines use a PROLED XDR display, with HDR peak brightness increased from 1,600 nits to 2,000 nits, replacing the notch with Dynamic Island, and bringing back the glowing Apple logo. There are currently no subsequent developments.

    4 sources32
  13. 16

    Neuralink team develops 8 brain-computer interface features in 72 hours

    In a 72-hour hackathon, the Neuralink team developed 8 functional prototypes for P21 participant Kenneth, including silently thinking "Hey Neuralink" in his mind and thinking through a task to trigger the Grok agent to operate a computer autonomously. The team also rewrote the neural voice activity detection system, achieving pure thought-based speech decoding without the need for button presses, and routed the decoded speech into the iPhone call microphone audio stream, allowing this ALS patient to call his wife using only his thoughts. In addition, the dedicated app Bloom supports placing orders by thought in a café, and a system-level brain-computer input method allows text entry in any app. On the same day, Elon Musk announced that Neuralink's first product is named Telepathy, stating that its output bandwidth will far exceed the fastest human speech or typing speed within a few years; the tweet cited a post about Neuralink engineers conducting a 72-hour hackathon with a speech-restoration research participant and rapidly prototyping multiple thought-communication features.

    4 sources32
  14. 17

    Microsoft offers voluntary severance packages to up to 60,000 employees

    On October 26, 2026, Microsoft offered voluntary severance packages to employees at level 70 and below across teams in the United States, Canada, India, and the United Kingdom, covering up to approximately 60,000 people, or about 27% of its global workforce; employees must decide by early December, and those who accept will leave the company by the end of the year. The Verge AI reported that this scale far exceeds the retirement plan offered to about 9,000 people earlier this year; as of June 30, Microsoft had approximately 223,000 employees, a decrease of 5,000 from the previous year. On October 29, Microsoft announced its financial results for the first quarter of fiscal year 2027, with Azure revenue up 48% year over year; the company said demand for AI infrastructure has exceeded available capacity; a few days before the earnings release, Microsoft offered voluntary departure packages to approximately 60,000 employees.

    19 sources31
  15. 18

    GitHub announces that Copilot Fusion is officially available

    On October 29, 2026, GitHub announced that Copilot Fusion is officially available. According to a report by Rohan Paul, the product can route coding tasks between Claude and GPT models; on Artificial Analysis's Coding Agent Index, its Max setting scored 60, compared with 62 for Claude Code, at a cost per task of $5.59 versus $11.42, with Fusion consuming 7.2M tokens per task, roughly half of Claude Code's usage. On the same day, Microsoft released multiple updates at GitHub Universe, stating that Copilot Fusion achieves frontier-level quality, is 40% faster, and reduces costs by up to 77%, and announced that a unified Copilot harness is entering Foundry as a managed service. There are currently no subsequent developments.

    6 sources29
  16. 19

    Altman Says New Model Solving Math Problems Excited Him, Then Left Him Sorrowful

    On October 30, 2026, OpenAI CEO Altman revealed in an interview that ChatGPT's new model, over an internal "Hold My Beer" weekend, successively cracked multiple unsolved interdisciplinary problems, with the proof related to the Navier-Stokes equations alone consuming $10 million in compute. He said the actual scientific discoveries produced far outnumber the several hundred results publicly disclosed, and admitted that the excitement was followed by more sorrow and melancholy, with a feeling that "something is coming to an end." On the same day, Altman responded to doubts from the mathematics community on the Cheeky Pint podcast, saying that the public's right to understand how the universe works outweighs the impact that releasing the results has on the mathematics community, and that the company needs to improve its communication but will not stop publishing, while acknowledging that its investment in math problems had at one point gotten out of control. Ayi AI Notes subsequently published an article, leaning toward believing that Altman's sorrow is genuine.

    2 sourcesInsufficient trend data29
  17. 20

    Reddit announces shift to a login-first experience

    Reddit announced it will change its nearly two-decade practice of allowing content to be read without an account, shifting to a login-first experience. CEO Steve Huffman said the platform will first test login requirements in markets outside the US. heise online reported that in Q3 Reddit had 130.3 million daily active users, of whom 76.8 million were logged out; Huffman said most unauthorized scraping comes from logged-out visits and stressed that registration does not require revealing one's real identity. Rohan Paul's report noted that Reddit will make login the default, after Google search referrals dropped by more than 1 million in a single week in October; in Q2, 30.1 million of the 53.2 million US daily active users browsed while logged out, and in Q3 US daily actives fell to 50.7 million, with Reddit also warning that daily actives could decline by a low double-digit percentage next quarter. The report also said Reddit stated that logged-in users are worth about three times as much in advertising value as logged-out visitors.

    4 sources28
How heat is calculated

Heat is a time-decayed index over a 48-hour window, counting each independent participant's best contribution once. It is not a count of articles, views, likes or participants. Compare values within the same ranking only. Missing intervals between real observations are interpolated; observed zeros remain on the baseline.