The Energy Efficiency of Agent Networks — VDF AI Benchmark White Paper

BENCHMARK WHITE PAPER v1.0 June 2026 VDF-WP-2026-003<br>The energy efficiency of agent networks.

A controlled benchmark of how VDF AI reduces the energy footprint of enterprise AI — by decomposing<br>work into DAG-based agent networks and dispatching each step through<br>SEEMR self-evolving model routing . The result: up to a 94.9% reduction in predicted<br>energy, with output quality held non-inferior in aggregate.

Authors VDF AI Research Team<br>Read time 16 min<br>License CC BY 4.0

Download PDF Read Online

ABSTRACT<br>Most of the energy an AI system consumes in production is spent at inference — the same<br>request answered again and again[6]. That energy is not a fixed<br>property of a model. It is the outcome of a decision: which model runs, broken into how many steps,<br>under what objective.

This paper reports a benchmark of that decision inside VDF AI. We compare a high-intensity baseline —<br>one large model answering the whole task — against two compounding strategies: routing each request<br>under an energy-aware objective, and decomposing a workload into a directed graph of smaller,<br>independently-routed stages. Across 71 configurations spanning four token budgets and five scenario<br>families, energy-led routing reduced predicted energy by 81–95% , with a stable<br>~94.8% reduction for the frontier-versus-compact pairing.

Crucially, savings without quality are meaningless. In a separate execution benchmark with a quality<br>score recorded per task, the routed condition reduced predicted energy by 94.9% while<br>remaining non-inferior in aggregate under a margin fixed in advance — with the<br>task-level exceptions disclosed in full. The contribution here is not a single number; it is an<br>auditable account of how routing and decomposition turn energy into something an enterprise can<br>measure, steer, and defend.

Keywords<br>energy-aware inference · DAG agent networks · self-evolving routing · non-inferiority · Green AI · token-level attribution · sustainable enterprise AI

AT A GLANCE<br>Six numbers from the benchmark

Peak energy avoided 94.9%<br>predicted energy removed by eco routing vs. a pinned frontier baseline

Efficiency multiple ≈20×<br>less predicted energy per workload at the same task, frontier vs. routed

Quality outcome Non-inferior<br>routed quality held within a pre-registered 0.10 margin in aggregate

Benchmark depth 71<br>configurations across five scenario families and four token budgets

Savings range 81–95%<br>reduction band observed across different model pairings

Selective frontier 54%<br>energy still avoided when one DAG stage deliberately keeps the frontier model

FIGURE 1<br>The same work, a fraction of the energy

Aggregate of the quality-constrained execution benchmark: a pinned high-intensity baseline versus<br>energy-aware routing, with the quality guardrail satisfied.

Pinned frontier baseline 3.80 Wh

VDF AI — routed 0.19 Wh

94.95% predicted energy avoided<br>≈20× more efficient, same task<br>±0.10 quality margin — held

Fig. 1. Predicted energy in watt-hours for an identical task set. Figures are<br>coefficient-based predictions under benchmark conditions, not measured wall power.

SECTION 1 Why inference energy is a decision, not a constant

A model is trained once and served billions of times. The integral of that serving tail now dominates<br>the one-off training spike[6][8], which<br>means the most leveraged place to reduce AI's footprint is the dispatcher that decides, per request,<br>which model runs and how the work is split.

Enterprises increasingly have to attribute that energy — for sustainability reporting, for<br>internal chargeback, and for procurement decisions that no longer accept a single annual number.<br>So the question this paper answers is concrete: if you hold the task fixed and change only the<br>routing and decomposition strategy, how much energy moves? And does quality survive the<br>change?

We answer with a benchmark rather than an assertion. Two forms of evidence are reported: a<br>coefficient-based comparison that isolates the effect of routing policy under fixed token assumptions,<br>and a quality-constrained execution benchmark that pairs each energy figure with a measured quality<br>score. The first tells us how big the lever is; the second tells us whether pulling it costs anything.

SECTION 2 · FIGURE 2 The routing objective is a dial you control

The same candidate pool, three presets. Eco leans into energy; Max-Quality deliberately holds the<br>heavy model. That Max-Quality lands at exactly 0% saving is the point — it proves the savings come<br>from the policy, not from a benchmark quietly favouring the small model.

Frontier-class vs. compact local model<br>Eco energy-led objective

94.8%

Balanced resolves to the same efficient pick here

94.8%

Max-Quality holds the frontier model by design

0%

Heavy tier vs. light tier<br>Eco narrower energy gap between candidates

81.4%

Balanced matches eco for this candidate set

81.4%

Max-Quality holds the heavy tier by design

0%

The...