✨ ultrathink ✨

TIL that the French Government spent taxpayer money in recreating @arena… and the results are as hilarious as you would expect 🤣

Introducing Nested Learning: A new ML paradigm for continual learning that views models as nested optimization problems to enhance long context processing. Our proof-of-concept model, Hope, shows improved performance in language modeling. Learn more: goo.gle/47LJrzI @GoogleAI

Baidu's Apollo Go is excited to celebrate our partnership with PostBus and the launch of the AmiGo autonomous mobility service in Switzerland in a local event dated November 5 in Altstätten in the canton of St. Gallen, with the presence of the first two autonomous AmiGo vehicles.

Flashcards and Quizzes are officially rolling out TODAY on the mobile app! You can customize the number of questions, difficulty, and topics all from the convenience of your phone. Because being a busy, popular socialite should never get in the way of your studies 😎

Here is Gemini CLI’s October 27th weekly update for v0.12.0 - 🎉 New Partner Extensions: - 🤗 Hugging Face extension: Access the Hugging Face hub. gemini extensions install github.com/huggingface/hf-mc… - Monday.com extension: Analyze your sprints, update your task boards, etc. gemini extensions install github.com/mondaycom/mcp - Data Commons extension: Query public datasets or ground responses on data from Data Commons. gemini extensions install github.com/gemini-cli-extens… - Model Selection: Choose the Gemini model for your session with /model. (abhipatel12) - Model Routing: Gemini CLI will now intelligently pick the best model for the task. Simple queries will be sent to Flash while complex analytical or creative tasks will still use the power of Pro. This ensures your quota will last for a longer period of time. You can always opt-out of this via /model. (abhipatel12) - Discussion: github.com/google-gemini/gem… - Codebase Investigator Subagent: We now have a new built-in subagent that will explore your workspace and resolve relevant information to improve overall performance. (abhipatel12 and silviojr) - Enable, disable, or limit turns in /settings, plus advanced configs in settings.json (silviojr) - Explore Extension: Users can now open the extensions page in their default browser directly from the CLI using the /extensions explore command. (JayadityaGit) - Configurable Compression: Users can modify the compression threshold in /settings. The default has been made more proactive (scidomino). - API Key Authentication: Users can now securely enter and store their Gemini API key via a new dialog, eliminating the need for environment variables and repeated entry. (galz10) - Sequential Approval: Users can now approve multiple tool calls sequentially during execution. (joshualitt) github.com/google-gemini/gem… 🧵

Now, Gemini’s Deep Research can pull in info from @Gmail, @GoogleDrive, and Chat when you connect your @GoogleWorkspace account to give you more context-aware reports. To try it, just select “Deep Research” in Gemini on desktop and choose your sources. Coming to mobile soon.

Our 7th gen TPU Ironwood is coming to GA!  It’s our most powerful TPU yet: 10X peak performance improvement vs. TPU v5p, and more than 4X better performance per chip for both training + inference workloads vs. TPU v6e (Trillium). We use TPUs to train + serve our own frontier models, including Gemini, and we’re excited to make the latest generation available to @googlecloud customers.

🚀 Hello, Kimi K2 Thinking! The Open-Source Thinking Agent Model is here. 🔹 SOTA on HLE (44.9%) and BrowseComp (60.2%) 🔹 Executes up to 200 – 300 sequential tool calls without human interference 🔹 Excels in reasoning, agentic search, and coding 🔹 256K context window Built as a thinking agent, K2 Thinking marks our latest efforts in test-time scaling — scaling both thinking tokens and tool-calling turns. K2 Thinking is now live on kimi.com in chat mode, with full agentic mode coming soon. It is also accessible via API. 🔌 API is live: platform.moonshot.ai 🔗 Tech blog: moonshotai.github.io/Kimi-K2… 🔗 Weights & code: huggingface.co/moonshotai

(1) Our team at @GoogleDeepMind has been collaborating with Terence Tao and Javier Gómez-Serrano to use our AI agents (AlphaEvolve, AlphaProof, & Gemini Deep Think) for advancing Maths research. They find that AlphaEvolve can help discover new results across a range of problems.

this is just bonkers, tbh.

If you are mad about water use by AI you should be *really* mad about water use by golf courses.

We’ve released an early preview of Qwen3-Max-Thinking—an intermediate checkpoint still in training. Even at this stage, when augmented with tool use and scaled test-time compute, it achieves 100% on challenging reasoning benchmarks like AIME 2025 and HMMT. You can try the current version in Qwen Chat and Alibaba Cloud API—more to come as training continues. Qwen Chat: chat.qwen.ai/?thinking=true Alibaba Cloud API （enable_thinking=True）: modelstudio.console.alibabac…

don't trust the EU.

Privacy Win in EU: Denmark Drops Mandatory Chat Control in CSAM Fight Danish Justice Minister Peter Hummelgaard announced Thursday that the EU Council Presidency will abandon mandatory detection orders in the Child Sexual Abuse Material (CSAM) proposal—marking a major defeat for mass surveillance advocates. Facing opposition from EU Member States and several stakeholders, Denmark now backs “voluntary” CSAM detection only, scrapping plans to force platforms to deploy technology that would have breached end-to-end encryption.   X celebrates the decision and appreciates the leadership from Poland, Italy and others to oppose mass surveillance across the EU. We will continue to monitor the progress of these negotiations and oppose any efforts to implement government mass surveillance of users.

an agent is most definitely not a workflow, a especially not an a priori or an imperative one.

The shift to on-device AI requires a higher privacy bar. We're introducing Provably Private Insights (PPI) to enable essential developer analysis of GenAI features with a verifiable guarantee that raw user data is never exposed. This system is a core component of building AI responsibly. PPI uses an open-source Gemma LLM inside a Trusted Execution Environment (Confidential Federated Analytics) to categorize unstructured data and derive insights. Only aggregated, differentially private statistics leave the secure environment. We've released the code to allow external verification of the end-to-end privacy guarantees. Read more about its deployment in Recorder and our blueprint by Artem Lagzdin and Daniel Ramage: goo.gle/47De8XV Preprint: arxiv.org/abs/2510.21684

"one last ship, I promise bro, just one last feature". this is my team on a Friday. It is quite remarkable for one reason: we are in the production domain, where things have incentives to go slower. but everyone is just cranking out good and fast. we are living a tidal change.

New patch version of Gemini CLI v0.11.2 is out 🚀 Model routing is now enabled by default... what the heck is model routing? It means based on complexity of the request different models will now be used: • "hi 👋" ➜ 2.5 Flash • "refactor my codebase" ➜ 2.5 Pro What does this mean for you, the user? • 📉 Reduced latency for a snappier experience • 💰 Extended quotas by using the Pro model only for tasks that truly require it. This should help alleviate some of the capacity issues on the Pro model, while letting your Pro quotas go much further and extending your sessions. Easily disable model routing with /model to only use one model. Read more about model routing: github.com/google-gemini/gem…

Introducing Kimi CLI Technical Preview & Kimi For Coding! Kimi CLI powers your terminal: - Shell-like UI + shell command execution - Seamless Zsh integration - MCP support -Agent Client Protocol (now compatible with @zeddotdev) More features incoming!