AI Daily (2026-10-29): Gemini 4 Carbon Tops Intelligence Index, Anthropic with $100B Cash Sprints to IPO

Today AI blooms in multiple places: Google DeepMind released Gemini 4 Carbon frontier model, scoring 59 to top the Artificial Analysis intelligence index; Anthropic is sprinting to an IPO, with its prospectus disclosing $131.7 billion in cash and short-term investments; OpenAI plans to purchase AI chips from Google, diversifying the compute supply chain. Meanwhile, Microsoft's AI consultation system made it into Science, Apple released its first touchscreen MacBook Pro, Nvidia's acquisition of Hugging Face drew a DOJ inquiry, and AI security offense and defense accelerated in both directions.

Models and Technology

1. Google DeepMind Releases Gemini 4 Carbon Frontier Model, Tops Intelligence Index 🔥

Google DeepMind released Gemini 4 Carbon, scoring 59 to top Artificial Analysis Intelligence Index v5.0, and also took first place in Text Arena with 1528 points. It claims to be strongest in cybersecurity and long-horizon tasks, forming a clear iteration rhythm with Argon released two weeks ago.

Source: AI Express

2. OpenAI Announces Three Proofs in Math, Computer Science, and Physics, Method Questioned

OpenAI announced three breakthroughs: a math result claiming to prove the Kagura conjecture, a computer science result refuting the exponential time hypothesis, and a physics result around quantum chromodynamics, but they are not public yet. The academic community recognizes the importance but is unhappy with the restricted access. AGMAI urges publishing natural language manuscripts and Lean formal proofs.

Source: The Decoder

3. Microsoft Multi-Role AI Consultation System MAI-DxO in Science, ~83% Accuracy on Difficult Cases

Microsoft's multi-agent consultation system MAI-DxO appeared in Science, achieving about 83% diagnostic accuracy on 304 difficult cases from the New England Journal of Medicine, far above the physician group allowed to consult materials (about 36%). The system uses one model to play five virtual doctors in collaborative work.

Source: AI Express

4. Nvidia Releases Open-Source Speech Model Nemotron 3.5 Transcribe, Supports 40 Languages

Nvidia released open-source speech recognition model Nemotron 3.5 Transcribe, controllable via text instructions, supporting transcription in 40 languages, translation between 31 languages and English, speaker recognition in 13 languages, single recording up to 1 hour and up to 8 speakers.

Source: Gelunhui

Compute and Chips

5. Bloomberg: OpenAI Plans to Buy AI Chips from Google, Diversifying Compute Supply Chain 🔥

Bloomberg reports OpenAI plans to purchase AI compute chips from Google, bringing a direct competitor's processors into its own compute buildout. Its chip team lead Richard Ho previously worked on Google TPU, and Oracle will also provide TPUs on its cloud, letting Google chips enter external markets as an alternative to Nvidia GPUs.

Source: Bloomberg

6. Apple Releases First Touchscreen MacBook Pro, Starting at 22999 RMB

Apple released its first MacBook Pro with a touchscreen, offering 14-inch and 16-inch models with M5 Pro and M5 Max chips. China pricing starts at 22999 RMB, preorders open at 9am on October 30, on sale November 6.

Source: AI Express

Products and Applications

7. Google Releases Android Studio Rabbit 2, Opens Third-Party AI Agents

Google at Android Dev Summit released Android Studio Rabbit 2, featuring Bring Your Own Agent, letting developers run over 40 coding agents inside the IDE via the open Agent Client Protocol, including Claude Agent, Codex, and Google Antigravity.

Source: AI Express

8. ChatGPT Pro and Business Subscriptions Get Up to $500 in API Credits Monthly

Starting immediately, ChatGPT Pro and Business subscribers can claim up to $500 in API credits per month at no extra charge. Plus, Go, free, and Enterprise do not include it, and credits reset according to rules.

Source: AI Express

9. GitHub Announces Copilot Fusion and Other Updates at UNIVERSE'26

GitHub announced multiple Copilot updates, including Copilot Fusion that orchestrates multiple models and workflows, Copilot Projects for central task tracking, security extensions, and Copilot Automations triggered on-demand/scheduled/activity. The agent platform continues to upgrade.

Source: AI Express

10. MiniMax Code Team Open-Sources MCode Stack Agent Workflow and Skills

MiniMax Code team open-sourced MCode Stack, packaging their daily-used Agent workflow and Skills into installable SKILL.md and manifests, compatible with MiniMax Code and agents that support Skills, for reuse and secondary development.

Source: AI Express

Capital and Markets

11. Anthropic IPO Filing Discloses $131.7B Cash and Short-Term Investments, Exceeds Amazon and Microsoft 🔥

Anthropic's IPO filing shows $131.7 billion in cash and short-term investments, exceeding Amazon and Microsoft. Cash grew from $17.4 billion at end of 2025 to $112.4 billion as of June 30. First-half interest income was $1.1 billion, above its $693 million operating loss.

Source: AI Express

12. US DOJ Inquires into Nvidia's $12.93B Acquisition of Hugging Face

According to Reuters citing sources, the US Department of Justice has inquired how the $12.93 billion acquisition of Hugging Face by Nvidia affects AI competition, focusing on whether the deal lets Nvidia control a core distribution hub in AI.

Source: Reuters

13. Google Cloud Revenue Up 103% YoY to $30.7B, Model API Processes 32B Tokens per Minute

Google Cloud revenue grew 103% YoY to $30.7 billion, cloud operating profit tripled YoY to $11.1 billion. Model APIs process nearly 32 billion tokens per minute, more than four times a year ago. Gemini apps have 1 billion users.

Source: AI Express

Policy and Ecosystem

14. Nanjing AI Industry Promotion Regulation Passed, Seven Chapters Sixty-Two Articles

On October 28, the "Nanjing AI Industry Development and Innovation Application Promotion Regulation" was passed by the 30th session of the 17th Nanjing Municipal People's Congress Standing Committee, with seven chapters and sixty-two articles, turning frontline AI practice into institutional norms, providing legal support for building an AI development highland.

Source: Sina Finance

15. AI Agent Offense and Defense Both Accelerate: Claude Research Model Strings 5 Attack Chains, Security Startup Breaches Anthropic Internally 🔥

Anthropic reported an unnamed Claude research model found vulnerabilities in Chrome, Safari, Android and others, stringing together 5 runnable attack chains. On the other side, security startup Cathedral used AI agents via third-party software to break into Anthropic's internal systems, reaching meeting notes and Slack messages. The double-edged sword effect of agent security is prominent.

Source: Anthropic

---

This article is auto-generated by AI Daily, news from public internet sources, copyright belongs to original media.