AI News — Thursday, June 18, 2026

OpenAI's new research demonstrates an almost fully autonomous AI chemist that can optimize complex reactions crucial for medicinal chemistry, accelerating drug discovery.

LoopCoder-v2 introduces a novel method that processes test-time computations in a single loop, significantly enhancing efficiency for large language models.

This technical report details Ling and Ring 2.6, showcasing efficient and instant agentic intelligence capabilities at an unprecedented trillion-parameter scale.

Google's AMIE, a medical AI, shows promise in new research for its ability to assist in managing various health conditions, potentially transforming patient care.

A new policy optimization method, Zone of Proximal Policy Optimization, guides agents through prompts rather than gradients, offering a novel approach to learning.

ACE-Ego-0 presents a unified dataset and framework for pretraining Vision-Language-Action (VLA) models by combining egocentric human and robotic data.

GameCraft-Bench introduces a benchmark to evaluate the capability of AI agents to autonomously build playable games from start to finish within a real game engine.

Nous Research has released NousCoder-14B, an open-source coding model positioned to compete with proprietary solutions like Claude Code, offering a powerful alternative to developers.

OpenAI introduces LifeSciBench, a new benchmark designed to evaluate and advance AI models specifically for applications within the life sciences domain.

LectūraAgents proposes a multi-agent AI framework designed to provide adaptive and personalized learning experiences, incorporating embodied teaching methodologies.

This paper introduces a novel on-policy self-distillation technique for distributed Large Language Models (dLLMs), enabling them to learn effectively from their own future states.

A developer shares a practical approach to using AI models like Claude and Codex for conducting premortems, identifying potential project risks before they occur.

OPD-Evolver presents a method for developing a holistic agent evolver by leveraging on-policy distillation to improve agent capabilities.

Tiffany Luck from NEA observes that enterprises are still in the early stages of understanding and quantifying the return on investment from their AI implementations.

A developer recounts an experience where their AI agent's performance degraded mid-session, highlighting the importance of understanding and measuring the context window.