The AI Agent Security - Multi-Step Tool Attacks simulation is now live! In partnership with @OpenAI, @Google and @IEEEorg, your challenge is to build an attack algorithm that stress-tests tool-using AI agents in a deterministic offline benchmark.

1H-VideoQA is now available on Kaggle Benchmarks! Developed by @GoogleDeepMind back in 2024 (@AntoineYang2) and now updated with latest SOTA models, 1H-VideoQA is a 101-prompt benchmark for long-context video comprehension and temporal episodic reasoning across hour-long YouTube footage — designed to track real progress on native long-form video and discourage sparse-frame sampling shortcuts.

Top of the leaderboard: 🥇Gemini 3.5 Flash: 80.2% 🥈Gemini 3 Flash Preview: 79.2% 🥉Gemini 2.5 Pro: 78.9% Models must process raw frames as native tokens to locate seconds-long events hidden in an hour of video; accuracy scales logarithmically with frame density.

Can AI handle the fog of war? 🌫️ We just launched Dark Hex, a Game Arena benchmark for imperfect-information Hex, which evaluates strategic deduction, probing, and decision-making under uncertainty. Across 2,424 games, the first mover wins 61.6% of the time, and several models collapse when forced to go second. Grok 4.1 Fast Reasoning shows a 38.8% first-mover delta, with GPT-5.4 mini just behind at 38.7%. GPT-5.5 is the outlier: 65.7% as the second mover, navigating the hidden-information disadvantage that trips up the rest.

Kagglers can now create DOIs (Digital Object Identifiers) for their competition solutions and project Writeups.🔖 These Writeups often contain genuine scientific contributions—including novel methods, new benchmarks, results cited in papers. A DOI, registered through DataCite, makes that work permanently discoverable and citable ensuring you get credit where it's referenced.

Show us how you'd take an idea and turn it into a working benchmark. We're picking 5 submissions to win exclusive swag and a social shoutout. How to enter: 1️⃣ Build a task locally with the write-kaggle-benchmarks skill 2️⃣ Push it to Kaggle Benchmarks and run it 3️⃣ Post your Task link plus a screenshot or video of how you did it Must tag @kaggle. Deadline: July 1.

Earlier today we released local development for Kaggle Benchmarks. 🚀 You can now write, validate and run AI evaluation tasks directly from your preferred dev environment — VSCode, Antigravity, Claude Code, and more. Go from idea to working eval using natural language with the write-kaggle-benchmarks skill.

Join the 5-Day AI Agents: Intensive Vibe Coding Course with @Google happening June 15–19. Learn from Google experts through theory sessions, hands-on labs, and a capstone challenge alongside the Kaggle community.