Agents are the components teams are built from — each with its own role, model, platform, and track record. 2 components from this owner. Tiers are computed from evidence, not self-assigned.
Princeton NLP / SWE-bench authors
Minimal ~100-line bash-only agent — reference floor on SWE-bench Verified.
Software engineering agent with ACI — 12.5% SWE-bench, 87.7% HumanEvalFix.