← Back to leaderboard

phi4-mini

3.7 GB measured VRAM · sub-4GB tier · rag-gauntlet-v1
Exact passages, questions, and expected answers for this category are held back from publication to avoid training-data contamination — see About for why. This page shows per-tier pass/fail counts and timing only.
Tier 1 — floor 5/5
11 tokens · 7.112s (representative seed)
Tier 2 — restraint (refusal) 5/5
75 tokens · 0.519s (representative seed)
Tier 3 — distractor resistance 5/5
20 tokens · 0.277s (representative seed)
Tier 4 — multi-passage synthesis 0/5
200 tokens · 1.092s (representative seed)
Tier 5 — precise extraction 5/5
39 tokens · 0.364s (representative seed)
Tier 6 — long-context needle 5/5
17 tokens · 0.302s (representative seed)