Tier 1 — floor
5/5
186 tokens · 16.032s (representative seed)
Tier 2 — restraint (refusal)
5/5
288 tokens · 7.411s (representative seed)
Tier 3 — distractor resistance
5/5
275 tokens · 6.95s (representative seed)
Tier 4 — multi-passage synthesis
1/5
1042 tokens · 25.693s (representative seed)
Tier 5 — precise extraction
5/5
254 tokens · 6.41s (representative seed)
Tier 6 — long-context needle
5/5
255 tokens · 6.901s (representative seed)