Stateless Language Agents Match Top Baselines with Far Fewer Tokens

Researchers from Stanford, CMU and UW propose Stateless Language Agents, which match the strongest baselines while using up to 84% fewer tokens by separating stateful search from stateless agents.

2026-10-12 ~ 2026-10-13 · 2 related posts

• Stanford's Stateless Language Agents Match Best Kernel Baseline With 84% Fewer Tokens — StanfordUniversity · 2026-10-12
• Stateless Language Agents hit best results with 14.5x fewer tokens across 9 research tasks — RulinShao · 2026-10-13