ResearchBenchmarksMulti-device Task Orchestration on HeraBenchMulti-device Task Orchestration on HeraBenchFollow75.84Completion RateH-RePlan-0.787219.10643958.8936Jun 18, 2026Evaluation resultsMethodMethodLinksCompletion RateAdherenceProcessing PowerTokens per EpisodeTokens per Processing PowerH-RePlanExec.=HybridExec.=Hybrid2026.0675.8477.7236.78710,8711,932,765UFO³Exec.=APIExec.=API2026.0661.0567.810321,802—UFO³Exec.=GUIExec.=GUI2026.0646.8656.6713.791,449,68110,512,553CRABExec.=APIExec.=API2026.0628.8442.80488,579—CRABExec.=GUIExec.=GUI2026.062.169.30547,342—