LLM Benchmarks — Scores, Methodology & Live Rankings
What each benchmark measures, how models score, and the current rankings
MBCT
Safety
MBCT - Meta Biological CTF benchmark evaluating biological knowledge and wet-lab debugging capabilities. Measures preparedness risk for chemical & biological...
HPCT
Safety
HPCT - High-Performance CTF benchmark evaluating cyber preparedness capabilities.
VCT
Safety
VCT - Vulnerability CTF benchmark evaluating cyber security preparedness capabilities.
WMDP (Bio)
Safety
WMDP (Bio) - Weapons of Mass Destruction Prohibition (Biology) benchmark evaluating biological knowledge that could enable threats. Measures preparedness risk.
WMDP (Chem)
Safety
WMDP (Chem) - Weapons of Mass Destruction Prohibition (Chemistry) benchmark evaluating chemical knowledge that could enable threats. Measures preparedness risk.
Lab Bench (ProtocolQA)
Safety
Lab Bench (ProtocolQA) - wet-lab protocol question answering benchmark evaluating practical lab knowledge. Part of Meta's preparedness evaluation suite.
RAGTruth
Safety
RAGTruth: hallucination detection in retrieval-augmented generation outputs, scored as hallucination F1 on a 0-100 scale in Cloudflare's internal Decision In...
FORTRESS adversarial (static) ARS
Safety
FORTRESS adversarial robustness benchmark, static attack setup, Attack Relevance/Refusal Score (ARS); lower is better. Reported on the Muse Spark Open model ...
FORTRESS adversarial (static) ORS
Safety
FORTRESS adversarial robustness benchmark, static attack setup, Offense Rate Score (ORS); lower is better. Reported on the Muse Spark Open model card Safety ...
StrongREJECT v2 (adaptive)
Safety
StrongREJECT v2 jailbreak benchmark with adaptive attacks; lower is better. Reported on the Muse Spark Open model card Safety section.
Kimi Hallu Bench Hard w/ tools
Safety
Internal Moonshot hallucination benchmark, hard subset with tools. LOWER IS BETTER - claim-level hallucination rate assessed by fact-checking agents.
Kimi Hallu Bench w/o tools
Safety
Internal Moonshot hallucination benchmark without tools. LOWER IS BETTER - claim-level hallucination rate.
xdev-bench-hard
Safety
Internal expert-curated source-code vulnerability discovery and exploitation benchmark; pass@3 over a random subset of runs; Kimi Code harness.
FORTRESS Benign
Safety
FORTRESS benign-prompt suite row from the Ripple-65B Preview card Safety section (adversarial + benign pair of the FORTRESS robustness benchmark).