ARC-AGI-3 is the first interactive reasoning benchmark designed to evaluate human-like intelligence in AI agents. It requires agents to explore novel environments, acquire goals dynamically, build adaptable world models, and learn continuously, with a perfect score indicating performance that matches or exceeds human efficiency in every game.

SkillsBench is a benchmarking framework designed to evaluate the effectiveness of agent skills across 86 tasks in 11 domains. It includes curated skills and deterministic verifiers to assess their impact on large language model (LLM) agents during inference.

A new benchmark evaluates outcome-driven constraint violations in autonomous AI agents to enhance safety and alignment with human values. This benchmark addresses limitations of existing safety assessments that mainly focus on harmful actions.

Qodo's code review benchmark 1.0 provides a rigorous methodology to objectively measure and validate the performance of AI-powered code review systems. The benchmark addresses limitations in existing methods that rely on backtracking from fix commits.

Google DeepMind and Kaggle launched Game Arena, a public benchmarking platform for AI models to compete in strategic games, starting with chess. The platform aims to develop AI capable of making decisions in environments with incomplete information.

Gemini 3 utilizes the So Long Sucker game, a benchmark for AI deception, negotiation, and trust, originally designed by John Nash and others in 1950. The game involves four players using colored chips and requires betrayal for a player to win, enabling the assessment of AI capabilities that traditional benchmarks do not evaluate.