Home › Evidence › Records › eng-0134

eng-0134

window recorded
citable URL: https://halobench.com/records/eng-0134/ — this address never moves; the anchor /records/#eng-0134 keeps resolving
149.1 Whmean 128.03 W − 10.1 W idle floor (box-idle-all-empty) = 117.93 W attributable
window 2026-08-16T12:58:58Z → 2026-08-16T14:08:51Z · provenance recorded
method wall-meter · aihydra energy sensor via Home Assistant (counter-difference on sensor.hardware_ai_hydra_energy, raw ~10s-resolution history samples, linearly interpolated to the run-meta window) · no contention
per task 6.4834 Wh · 0.1964 p at 30.3 p/kWh
covers run-0345
aihydra ornith-35b-fullbench tau2 26-task run window (12:58:58Z-14:08:51Z, 4193s), matching run-0345 exactly (excludes the preceding 13s server-load window, 0.1246 Wh / 34.5 W, per the deepseek-v4-flash/laguna-s-21 convention -- including it moves the figure by <0.1%: 149.2420 Wh total, 6.4888 Wh/correct). Counter interpolated from raw ~10s-resolution HA history samples pulled fresh this session: start 22.0771553905 kWh, end 22.2262727988 kWh. wh_per_task is Wh-per-CORRECT-ANSWER: wh_total / tasks_passed (23 of 26, run-0345.metrics) = 6.4834 Wh, 0.1964 pence at the standing 30.3 p/kWh tariff. WHOLE-SESSION energy -- includes simulator (user_llm) wait time end to end, not an agent-only or decode-only figure, directly comparable to deepseek-v4-flash's eng-0107, nemotron3-super's eng-0116 and laguna-s-21's eng-0124 on the same denominator. THIS IS THE NEW LEADER among full standard-26-task tau2 comparisons on this box, ahead of laguna-s-21's 7.2588 Wh, deepseek-v4-flash's 24.29 Wh, nemotron3-super's 32.43 Wh and qwen38-27b's 38.29 Wh -- screened against every other benched model's Wh-per-correct-answer before being claimed (the qwen35-122b 5.71 Wh/correct figure is a 5-task SMOKE window predating the standard-26-task convention and remains not a like-for-like comparator, same caveat laguna-s-21's page already carries).

Wh, never joules — read as a cumulative-counter difference across the window; delta over the idle baseline is the only figure that means anything

Cited by — computed at build time, never stored