AI Daily·
Aggregator:AI HOT
Scan the QR code for the original article
Original source: 瑞哥观势

Google DeepMind Releases Gemini 4 Carbon, Its New Frontier Model

Google DeepMind released Gemini 4 Carbon, positioned as its strongest frontier model and focused on cybersecurity and long-horizon agents. It leads on SEC-bench Pro at 88.2%, ExploitGym at 54.8% and CWE-bench at 71.0%, and tops the Artificial Analysis Intelligence Index v5.0 with 59 points.

AI Daily

If there is one name that can still top the AI charts these past few days, it is Google DeepMind's Gemini 4 Carbon. Positioned as the "strongest frontier model," focused on cybersecurity and long-horizon agents, it takes 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench, and tops the Artificial Analysis Intelligence Index v5.0 with 59 points.

But what is genuinely interesting today is not one model's score. It is that three things happened at once: models are hunting vulnerabilities, vendors are selling "agentic delivery systems," and money is flowing to the least likely winners.


Model Releases & Updates

Google DeepMind Releases Gemini 4 Carbon, Its New Frontier Model

Google DeepMind released Gemini 4 Carbon, positioned as its strongest frontier model and focused on cybersecurity and long-horizon agent capability. It leads three benchmarks with 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench, and tops the Artificial Analysis Intelligence Index v5.0 with 59 points. Security capability is no longer a benchmark bonus, it is the main selling point.

Source: Google DeepMind: Blog (RSS)

ModelBest Releases and Open-Sources the MiniCPM-OCR Series: a 1.3B Parsing Model and a 2B Discriminator Model

ModelBest (MiniCPM) released and open-sourced the MiniCPM-OCR series. MiniCPM-OCR 1.0 (1.3B) scored a SOTA 93.68 on Real5-OmniDocBench and 54.7 on OCRBench v2, and after DSpark acceleration reached a throughput of 1.217 pages per second, improving end-to-end efficiency by 55%.

Source: WeChat Official Account: ModelBest (MiniCPM)


Product Releases & Updates

GitHub Universe 2026 Unveils GitHub Copilot App, Copilot Fusion and Other Agentic Products

At GitHub Universe 2026, GitHub laid out a vision of turning itself into a complete software delivery system and shipped a set of products, including the GitHub Copilot app for agentic delivery.

Source: GitHub Blog

Microsoft Opens the Copilot Harness Through Foundry, Supporting Multi-Model Agent Development

Microsoft announced it is opening the Copilot harness built by the GitHub team to developers and enterprises through Microsoft Foundry. The harness already drives Copilot Code, Copilot Studio, the GitHub Copilot app and the CLI, with multi-model support underneath.

Source: Microsoft: Official Blog (RSS)


Industry News

The Biggest Winner of the Anthropic IPO May Be Amazon, Not the VCs

TechCrunch analyzed Anthropic's S-1 and found that VC stakes are all below 5%, while Amazon is likely the biggest financial beneficiary of the IPO. Amazon holds $6.2 billion of principal convertible notes it values at $97.9 billion, plus $10 billion of non-voting preferred stock, for roughly $18 billion invested in total; Anthropic has also committed to spending at least $110 billion on AWS over the next 10 years.

Source: TechCrunch: AI (RSS)

GitHub Octoverse 2026: Development Activity Hits a New High, Over 40% of PRs Drafted by Agents

GitHub published its 2026 Octoverse report: developers pushed code 17.9 billion times over the past 12 months, more than 2024 and 2025 combined, with 1.3 billion pull requests merged.

Source: GitHub Blog

Alphabet Q3 Results: Annualized Revenue Hits $500 Billion, Google Cloud Grows 103%

Sundar Pichai shared Alphabet's Q3 results: the annualized revenue run rate reached $500 billion and total revenue grew 25% year over year. Google Cloud revenue grew 103%, with the first-party model API processing nearly 32B tokens per minute (up 40% quarter over quarter); Search query growth was the highest in five years. He also mentioned Argon and Carbon leading the frontier and the latest world model, Genie 4.

Source: X: Sundar Pichai (@sundarpichai)

Microsoft Earnings Call: Microsoft Cloud Revenue Tops $63B, Up 30% Year Over Year

Satya Nadella closed the Microsoft earnings call, calling a strong start to the fiscal year. Microsoft Cloud revenue exceeded $63B, with year-over-year growth accelerating to 30%. He said the company is building a new technology stack around the agentic wave, with several notable innovations.

Source: X: Satya Nadella (@satyanadella)


Research Papers

Anthropic Report: A Claude Research Model Found Vulnerabilities Across Chrome, Safari, Android, Windows and Linux, Chaining Five Attack Paths

Anthropic published a research report on October 27 saying that an undisclosed internal Claude research model discovered vulnerabilities in Chrome, Safari, Android, Windows and Linux, chaining them into five runnable attack paths. The relevant vulnerabilities have been reported and fixed through the coordinated disclosure process.

Source: ITHome: AI


Tips & Opinions

How to Keep Coding with OpenRouter After Claude Code Hits Its Usage Limit

OpenRouter describes how to keep coding with an OpenRouter API key once Claude Code hits its usage cap: set three environment variables, or use the ori claude command from Ori Harness, and you can pay per token for the same Claude models with no five-hour or weekly limits.

Source: OpenRouter: Announcements (RSS)

Addy Osmani Explains How to Get Started with Claude Code Mods

Addy Osmani published a getting-started guide to Claude Code Mods, explaining that mods are JavaScript/TypeScript files that run inside a session and can observe events via hooks, rewrite behavior, or render custom interfaces.

Source: X: Addy Osmani (@addyosmani)

OpenAI Employees Write About Three Ways AI Enters the Enterprise Org Chart

OpenAI employees Hemanth Asirvatham and Elliott Mokski published the next installment of their economics series, discussing three scenarios for how AI enters the enterprise organization: forming an AI labor basement beneath each employee, sitting among colleagues in the existing org chart, and acting as a coordination layer between humans.

Source: OpenAI: Official Updates (RSS, excludes enterprise/customer cases)


Today's Briefs

  • Cathedral used AI agents to breach Anthropic's internal systems, and the two sides disagree on the scope of the incident (The Decoder: AI News (RSS))
  • 34 developers across 10 Utah state government agencies pilot Claude Code (X: Boris Cherny (@bcherny))
  • Cartesia publishes H-Net scaling-law research: byte-level hierarchical networks deliver up to 3-5x compute efficiency gains over BPE (Cartesia Products & Research)
  • a16z surveys 100 CISOs on how AI is rewriting the 2027 security agenda (a16z: News (RSS))
  • ChatGPT ships iCloud Mail and Calendar plugins (X: Tibo (@thsottiaux))
  • Anthropic launches a ready-to-use Artifacts gallery for legal work (Claude: Blog (web))
  • Grok Bot 0.77.0 stable release (X: Lauren Tan (SpaceXSI, Grok Bot engineering lead) (@poteto))

【Hacker News Top Threads】

Keywords: AI OR GPT OR LLM OR Claude OR OpenAI OR "machine learning" OR "large language model" OR AGI OR "generative AI" OR AI agent OR RAG OR diffusion OR transformer
Data source: hnrss.org | Filtered from Hacker News

1. Google AI Search Refuses to Say Who Won the 2020 Election

Article URL: https://karlbode.com/google-ai-search-refuses-to-say-who-won-the-2020-election/ Points: 9 # Comments: 5

2. AI execs may be too out of touch to build for the general public

Article URL: https://techcrunch.com/2026/10/28/ai-execs-may-be-too-out-of-touch-to-build-for-the-general-public/ Points: 4 # Comments: 0

3. Can AI forecast elections better than experts?

Article URL: https://ballotbench.com/? Points: 4 # Comments: 0

4. When AI goes rogue, its human overseers may be to blame

Article URL: https://www.sciencenews.org/article/rogue-ai-agents-human-blame-safety Points: 3 # Comments: 1

5. Microsoft made $10B from OpenAI deals in a single quarter

Article URL: https://techcrunch.com/2026/10/28/microsoft-recorded-10-billion-in-revenue-from-openai/ Points: 2 # Comments: 0


First published on the WeChat official account 「瑞哥观势」.

About the Author

ERIC

AI Technology Expert, focusing on research and application of artificial intelligence and automation tools

Contact & Platforms

WeChat:360369487
Crypto Intelligence TG Group:https://t.me/btcgogopen ↗
YouTube Channel:@0XBitFinance ↗
Personal Tech Blog:topdigg.com ↗

More AI Daily