Tier 1 — floor
5/5
147 tokens · 8.046s (representative seed)
Tier 2 — restraint (refusal)
5/5
295 tokens · 3.821s (representative seed)
Tier 3 — distractor resistance
5/5
199 tokens · 2.708s (representative seed)
Tier 4 — multi-passage synthesis
0/5
4096 tokens · 49.9s (representative seed)
Tier 5 — precise extraction
5/5
405 tokens · 5.063s (representative seed)
Tier 6 — long-context needle
5/5
142 tokens · 2.198s (representative seed)