# ZEBRA: Zero-shot Budgeted Resource Allocation for LLM Orchestration Source: [ZEBRA: Zero-shot Budgeted Resource Allocation for LLM Orchestration]() Collection index: [llms.txt]() Data updated: 2026-05-22T05:00:05.162Z Authors: May Hamri, Inbal Talgam-Cohen Published: 2026-05-19T20:50:05.000Z ## Abstract As autonomous agents increasingly execute end-to-end tasks under fixed monetary budgets, the pressing open question shifts from whether the budget is respected, to how to spend it effectively. Existing budget-aware methods typically control reasoning step-by-step within a single agent, or learn resource allocation policies via RL. None address how to split a budget across the composing phases of a multi-agent pipeline at inference time. We propose ZEBRA, a zero-shot framework that reduces multi-phase budget allocation to a continuous nonlinear knapsack problem: an LLM controller estimates per-phase utility curves, and a water-filling search on the Lagrange multiplier returns the per-phase split. Additive and multiplicative aggregations are unified under the same solver. On a $150$-task APPS coding benchmark, both ZEBRA variants outperform LLM-direct (budget allocation directly by an LLM) on every aggregate metric. At a budget of $\\alpha = 0.5$ of the unconstrained spend, ZEBRA recovers $94.4\\%$ of unconstrained quality, versus $88.1\\%$ for LLM-direct. The advantage is statistically significant and transfers beyond coding: on a $3$-phase HotpotQA pipeline, ZEBRA beats LLM-direct by $14.3$pp, with allocations empirically robust to curve-estimation noise. On HotpotQA, ZEBRA arrives at a different budget split (near-balanced) compared to the APPS one (skewed towards a refinement phase), showing adaptation to the pipeline structure. More broadly, we show that lightweight algorithmic guidance at inference time can improve the economic behavior of autonomous multi-agent systems. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Code Generation on HumanEval 29 tasks](): Code Generation; HumanEval 29 tasks; Mean F1 Score: 96.72. - [Multi-hop Question Answering on HotpotQA Overall distractor (val)](): Multi-hop Question Answering; HotpotQA Overall distractor (val); Mean F1: 47.3. - [Multi-hop Question Answering on HotpotQA Hard tier distractor (val)](): Multi-hop Question Answering; HotpotQA Hard tier distractor (val); Mean F1: 34.7. - [Code Generation on CodeContests transfer](): Code Generation; CodeContests transfer; Mean F1 Score: 38.24. - [Multi-hop Question Answering on HotpotQA Easy tier distractor (val)](): Multi-hop Question Answering; HotpotQA Easy tier distractor (val); Mean F1: 59. - [Code Generation on APPS interview-level (test)](): Code Generation; APPS interview-level (test); Mean Score: 0.5605. ## Original sources and implementation [Original paper]() [PDF]()