AI RESEARCH

Cross-Lingual Token Arbitrage: Optimizing Code Agent Context Windows via Local LLM Preprocessing

arXiv CS.AI

ArXi:2606.03618v1 Announce Type: new AI-assisted coding agents are bottlenecked by input-token cost. Two pathologies of raw human input drive much of this overhead: tokenization inefficiency for non-English text and structural entropy in conversational prompts. Existing approaches act reactively by compressing already-bloated contexts or intervening after failures occur. We evaluate on OMH-Polyglot, a multilingual coding benchmark spanning Turkish, Arabic, Chinese, and code-switched specifications.