UCLA 研究称 AI 智能体输出约 18% 的 token 属可删冗余
UCLA 与东北大学联合论文提出一种代理式审计方法：让 LLM 裁判标记智能体输出中可安全删除的片段，再重跑任务验证。在 6 个前沿模型与 4 个基准上，约 18% 的输出 token 属冗余；对 Kimi-K3 在 SWE-bench Verified 上剪除重复与填充内容后，token 减少 41%，成功率仍保持 89.5%（原为 89.0%）。结论是降本应定位任务与执行框架中的浪费来源，而非简单限制 token 上限。
报道正文
Rohan Paul (@rohanpaul_ai) · 2026-11-09 12
New UCLA+ Northeastern University paper finds that about 18% of AI agent output tokens are removable waste, and cutting it barely hurts task success.
Agent bills are usually measured as total tokens. That number can't tell you which tokens did useful work and which just repeated a policy, refetched a file, or went down a dead end.
They route agent calls through a proxy and have an LLM judge mark spans that can go without losing useful work. Then they delete those spans and rerun the agent to check nothing breaks.
Across 6 frontier models and 4 benchmarks, about 18% of output tokens were waste. Pruning repetition and filler from Kimi-K3's reasoning removed 41% of it on SWE-bench Verified, and success stayed at 89.5% versus 89.0%.
To cut agent costs, find where your task and harness create waste instead of just capping tokens.
