Context Language Models

Researchers introduced Context Language Models (CLMs), which allow language models to independently manage their own context by treating it as an editable file rather than relying on external control systems. This approach significantly improves performance across multiple tasks—achieving 11.4% higher accuracy with fewer computations on BrowseComp-Plus and 47.6% improvement on another benchmark—while enabling both natural language steering and reinforcement learning optimization of context management strategies. The method also introduces computational efficiencies for serving these models, reducing server-side compute by 35% compared to standard approaches.