Contrastive Language Models

Researchers introduced Contrastive Language Models (CLMs), a new class of AI model that uses contrastive learning to connect states and actions for fast decision-making. CLM-8B achieves performance comparable to existing models like Jev on computer-use, gaming, and tool-calling tasks while running up to 9× faster, and sets new benchmarks on coding tasks with lightweight fine-tuning. The model was trained on 60M Q&A pairs, 30M synthetic hard negatives, and 1M agentic trajectories, with established scaling laws showing predictable performance improvements based on compute and dataset size.