UCLA and Northeastern paper finds 18% of agent tokens removable
The non-obvious takeaway is that token waste is a property of the task and harness, not the model, so harness design changes may cut costs more than switching to a cheaper model. Engineering teams running coding agents or long multi-step workflows can use the proxy-and-rerun method to audit their own traces. The Kimi-K3 result on SWE-bench Verified shows a 41% token cut with success intact, a concrete benchmark for anyone comparing agent efficiency across frontier models.
A paper from UCLA and Northeastern University researchers reports that about 18% of AI agent output tokens are removable waste, and cutting them barely hurts task success. Agent bills are usually measured as total tokens, a number that cannot distinguish tokens doing useful work from those repeating a policy, refetching a file, or following a dead end. The finding gives teams a way to audit where agent spending actually goes.
The method routes agent calls through a proxy, then has an LLM judge mark output spans that can be deleted without losing useful work. Those spans are removed and the agent is rerun to verify 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 its tokens on SWE-bench Verified while success stayed at 89.5% versus 89.0% unpruned.
The practical implication is that teams looking to cut agent costs should identify where their task and harness create waste rather than simply capping token budgets. The full paper is available via the linked X post by Rohan Paul.
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.
This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include agents, Kimi-K3, SWE-bench Verified, token efficiency.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai).
