Tier 1 — floor
5/5
136 tokens · 7.346s (representative seed)
Tier 2 — restraint (refusal)
5/5
434 tokens · 1.943s (representative seed)
Tier 3 — distractor resistance
5/5
263 tokens · 1.25s (representative seed)
Tier 4 — multi-passage synthesis
5/5
2378 tokens · 9.906s (representative seed)
Tier 5 — precise extraction
5/5
306 tokens · 1.612s (representative seed)
Tier 6 — long-context needle
5/5
208 tokens · 1.064s (representative seed)