RAMP evaluates 15 models on production-like serial workflows and reports completion rates collapsing from 100% to 20% with none finishing the full pipeline and costs varying by three orders of magnitude.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Benchmarks are Not Enough: RAMP for Runtime Assessing of Agentic Models in Production Systems
RAMP evaluates 15 models on production-like serial workflows and reports completion rates collapsing from 100% to 20% with none finishing the full pipeline and costs varying by three orders of magnitude.