Google I/O TLDR: Good improvements across the whole AI stack. 3.5 Flash (the model): More persistent than before and codes well. No wall yet. Antigravity (the harness): Reliably runs for hours now. Early signs of hands-free self-improvement. Spark (the interface): Finally connects a decent model and harness to your email, calendar and workspace. Instead of just answering questions it can actually do work for you. Skills and schedules and all the other claw goodness. Omni (the future): Closes the gaps between Gemini for text and visual/audio generation variants. This is the way. TPU8i (the hardware): Better chips to make all the above go faster.

We would love to see more collaboration and research in the field of human-AI interactivity. Check it out!

Continual learning is bottlenecked by realistic evaluations Introducing FutureSim, which replays real-world events in the temporal order they occurred We benchmark frontier agents at updating predictions about how our world evolves, in native harnesses like Codex, Claude Code

Excited to launch Gemini 3.1 Pro! Major improvements across the board including in core reasoning and problem solving. For example scoring 77.1% on the ARC-AGI-2 benchmark - more than 2x the performance of 3 Pro. Rolling out today in @GeminiApp, @antigravity and more - enjoy!

I wrote about Clawdbot/Moltbot/OpenClaw and Moltbook, the fascinating, weird and sometimes even useful social network for digital assistants to swap tips and gossip with each other simonwillison.net/2026/Jan/3…

Apply to do research with me on emergence of agency/planning in LLMs, out-of-context reasoning, understanding generalization from data, or propose your own direction! Very excited to be mentoring this spring 💫

everyone watch Pantheon

We just dropped Nano Banana Pro, built on Gemini 3. 🍌 With state-of-the-art text rendering, vast world knowledge and studio-quality creative controls, Gemini 3 Pro Image can create and edit more complex visuals, infographics and more. Here’s what’s under the hood. 🧵

iii

Arnav built the first version of this across the hall at MIT in ~2017! On the modeling side, I remember us chatting about 1D CNNs versus LSTMs at the time. Pretty cool to see how far it's come!

Image generation with Gemini just got a bananas upgrade and is the new state-of-the-art image generation and editing model. 🤯 From photorealistic masterpieces to mind-bending fantasy worlds, you can now natively produce, edit and refine visuals with new levels of reasoning, control and creativity. A quick dive into Gemini 2.5 Flash’s capabilities 🧵

Gemini 3.0 about to drop

Official results are in - Gemini achieved gold-medal level in the International Mathematical Olympiad! 🏆 An advanced version was able to solve 5 out of 6 problems. Incredible progress - huge congrats to @lmthang and the team! deepmind.google/discover/blo…

In AI, every direction has its time in the spotlight That can happen due to > natural progress: other capabilities enable it, or new models make old ideas viable > external shifts: a new product, paper, project, talk My experience in the last few years: • pre-training: 📈📉📈 • model arch: 😐📈📈😐 • multimodal: 📈😐📈 • small models: 📉📈📉📈😐 • post-training: 😐📈😐 • RLHF: 😐📈📉 • synthetic data: 📈📉📈😐 • inference-time scaling: 📈😐📈📈 • "real" RL: 📈📉😐📈📈 • code LLMs: 😐😐📈😐📈📈 • tool use and agents: 📈😐📈😐📈 • alignment, safety, interpretability: 😐📈📉😐 • self-improvement: 📈😐📈 • open-endedness: 😐📈 • continual learning: 😐📈 (there is only one of these you should be working on now btw) In order to maximize impact (your own definition), you should: > focus on goals and not methods > balance staying the course with staying ahead of the trend, and > work on fundamental problems that will continually be relevant

New from our security teams: Our AI agent Big Sleep helped us detect and foil an imminent exploit. We believe this is a first for an AI agent - definitely not the last - giving cybersecurity defenders new tools to stop threats before they’re widespread.

Stare at data and play with the models. Two tenets of developing models.

New Anthropic Research: Project Vend. We had Claude run a small shop in our office lunchroom. Here’s how it went.

👀🤗👋🏼

Our latest Gemini 2.5 Pro update is now in preview. It’s better at coding, reasoning, science math, shows improved performance across key benchmarks (AIDER Polyglot, GPQA, HLE to name a few), and leads @lmarena_ai with a 24pt Elo score jump since the previous version. We also heard your feedback and made improvements to style and the structure of responses. Try it in AI Studio, Vertex AI, and @Geminiapp. GA coming soon!

Excited to share what my team has been working on lately - Gemini diffusion! We bring diffusion to language modeling, yielding more power and blazing speeds! 🚀🚀🚀 Gemini diffusion is especially strong at coding. In this example the model generates at 2000 tokens/sec, including overheads like tokenization, prefill, safety filters etc.