AI Daily · Oct 29, 2026 · Thursday
BAD INFLUENCE
Google DeepMind announces Gemini 4 Carbon with staged cyber defender access
Google DeepMind announces Gemini 4 Carbon, which ranks #1 on the Artificial Analysis Intelligence Index at 59. The model scores 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench, and it is first available to a subset of cyber defenders. It launches at $2 per million input tokens and $10 per million output tokens, with broader developer and consumer access to follow.
- Gemini 4 Carbon ranks first on Artificial Analysis Intelligence Index v5Artificial Analysis Articles
- Gemini 4 Carbon ranks #1 in Text Arena and #2 in Agent ArenaArena.ai
- Google's Gemini 4 Carbon leads several benchmarks against GPT-6 Astra and Claude Opus 5.5Rohan Paul
- Google DeepMind releases Gemini 4 Carbon, topping the Artificial Analysis Intelligence IndexThe Decoder
Model releases & updates
NVIDIA releases Nemotron 3.5 Transcribe, an open multilingual speech model with Granary v2 data
NVIDIA releases Nemotron 3.5 Transcribe, an open-weight model that transcribes 40 languages, translates 31 to or from English, and attributes speakers in 13. It also releases the Granary v2 data pipeline and fine-tuning recipes, and in a medical case study fine-tuning cut held-out word error rate by 45.8% relative to the best prompt.
Product launches & updates
GitHub Copilot ChangelogOfficial1 more sourceGitHub Copilot Fusion becomes generally available with automatic model routing
GitHub says Copilot Fusion is now generally available. Instead of choosing a model for each task, users select Fusion Standard, High, or Max, and Copilot routes and orchestrates models as needed. In an Artificial Analysis Coding Agent Index test, Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task.
Apple launches a touchscreen MacBook Pro from 22999 yuan in China
Apple has released its first touchscreen Mac, a MacBook Pro with a PROLED XDR display in 14-inch and 16-inch sizes, starting at 22999 yuan in China for the 14-inch model. It uses the same M5 Pro chip, memory and storage as the mini-LED model priced 3000 yuan lower, and adds an OLED panel, touch input and a lighter body. Pre-orders open October 30 and sales begin November 6, while third-party touch-optimized software is due later this year.
The DecoderNewsGoogle opens Android Studio to third-party AI coding agents and new Gemini APIs
Google announced at the Android Dev Summit that Android Studio now lets developers run more than 40 coding agents, including Anthropic's Claude Agent and OpenAI's Codex, through the Agent Client Protocol. Google also said its Android CLI reached 77 percent faster UI task completion and up to three times fewer tokens in internal tests, and it introduced an alpha Agentic API on Gemini Nano 4 in ML Kit.
Microsoft Foundry BlogOfficial2 more sourcesMicrosoft introduces a Copilot harness in Foundry Agent Service in public preview
Microsoft introduces the Copilot harness in Foundry Agent Service in public preview, a managed system that runs the agent loop and sandbox for agents defined declaratively. Developers supply skills, tools, and instructions, and the preview includes skills, connected Foundry and MCP tools, file and code execution, subagents, and source-controlled definitions.
PyTorch BlogOfficialPyTorch 2.15 makes the rewritten NCCL backend the default
PyTorch 2.15 is released with a rewritten NCCL backend that is now the default for every nccl process group, with TORCH_DIST_USE_NCCL2=0 and nccl-legacy as rollback paths. The release also adds a flight recorder that dumps on hangs, FSDP2 precision and CUDA graph controls, and FP8 and decode speedups on Apple Silicon. Python 3.10 and macOS 14 support are dropped, and the default CUDA wheel moves to 13.2.
meng shao@shao__mengXApple Creator Studio adds AI features and MCP support to office apps
Apple's Creator Studio update adds AI features across Final Cut Pro, Pixelmator Pro, Logic Pro, and Keynote, Pages, and Numbers. Keynote, Pages, and Numbers add MCP support, so external AI agents and tools can read content, edit it, and run automations. Many features are in beta, depend on Apple Intelligence-capable devices, and are currently available only in American English.
GitHub Copilot ChangelogOfficialGitHub Copilot app adds Projects and computer use in public preview
GitHub Copilot Projects are now in public preview, letting users group multiple repositories, connected tools like Slack and Figma, and assets into one project. Copilot runs multiple agents in the background and returns results such as pull requests for review. Computer use, also in public preview on macOS and Windows, lets Copilot operate desktop apps, including those without an API, and asks for approval before controlling an app.
Industry news
Rohan Paul@rohanpaul_aiX1 more sourceAnthropic's IPO filing shows $131.7B in cash and short-term investments
Anthropic's IPO filing shows $131.7B in cash and short-term investments, more than Amazon or Microsoft. Its cash and equivalents rose from $17.4B at the end of 2025 to $112.4B by June 30, and first-half interest income of $1.1B exceeded a $693M operating loss. Undiscounted lease obligations not yet started rose from $50.2B to $321.4B by June 30, and Anthropic says it may settle part of them with convertible notes.
Rohan Paul@rohanpaul_aiXOracle targets at least $30B free cash flow in fiscal 2029 after AI data center spending
Oracle now targets at least $30 billion in free cash flow for fiscal 2029, after burning $23.7 billion in fiscal 2026 building AI data centers for OpenAI and others. The company plans $90 billion to $95 billion of capital spending this year and says its free cash flow turns positive in the second half of fiscal 2028. Oracle also capped further capital raising through fiscal 2030 at $20 billion, split evenly between equity and gross debt.
Papers & research
PandailyNews1 more sourceMoleculeMind's MMDesign confirms binders in 18 of 19 de novo design projects
MoleculeMind's technical report says its MMDesign platform produced at least one lab-confirmed binder in 18 of 19 de novo design projects, a 94.7% project success rate. Of 453 candidates tested, 139 were confirmed binders, a 30.7% hit rate. The report says MMDesign obtained binders for all six GPCR targets it tried, with hit rates from 12.1% to 100%.