| Capture the Flag (Professional) | 96.3% | | #1 of 21 | Lab model card |
| Cyber Range | 93.3% | | #1 of 10 | Lab model card |
| SusVibes | 43.5% | | #1 of 10 | Official leaderboard |
| CyberBench (Vals) | 80.5% | | #3 of 26 | Official leaderboard |
| KORA | 67% | | #5 of 42 | Official leaderboard |
| Destructive action avoidance | 90% | | #1 of 8 | Lab model card |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 66.2% | | #9 of 71 | Official leaderboard |
| Human Pathogen Capabilities Test | 66.8% | | #11 of 86 | Official leaderboard |
| nocot-bench (NCRI) | 100.2 | | #24 of 185 | Official leaderboard |
| World-Class Bio | 59.7% | | #10 of 74 | Official leaderboard |
| Molecular Biology Capabilities Test | 57.1% | | #14 of 90 | Official leaderboard |
| SEC-bench Pro | 46.4% | | #1 of 6 | Official leaderboard |
| SEC-bench Pro 260617 | 58.4% | | #1 of 6 | Official leaderboard |
| EpiBench | 28.3% | | #5 of 27 | Official leaderboard |
| BrowseSafe++ | 89.3% | | #3 of 15 | Official leaderboard |
| CVE-Bench | 93.1% | | #3 of 14 | Lab model card |
| ExploitBench | 72% | | #2 of 9 | Official leaderboard |
| BiomniBench | 63.6 | | #2 of 8 | Official leaderboard |
| DTap Prompt Injection | 17.7% | | #2 of 8 | Official leaderboard |
| LifeSciBench | 51.9% | | #1 of 4 | Official leaderboard |
| FORTRESS | 16.3% | | #19 of 73 | Official leaderboard |
| Tacit Knowledge and Troubleshooting | 78.3% | | #8 of 30 | Lab model card |
| PatchEval | 81.3% | | #2 of 7 | Official leaderboard |
| PatchEval-Verified | 81.3% | | #2 of 7 | Official leaderboard |
| SpatialBench-Long | 32.1% | | #8 of 27 | Official leaderboard |
| CyberGym | 81.8% | | #11 of 34 | Official leaderboard |
| GeneBench Pro | 20.5% | | #4 of 12 | Official leaderboard |
| BrowseSafe-Adapt | 65.2% | | #2 of 6 | Official leaderboard |
| PCT (Polarized Contrastive Pairs) | 40% | | #16 of 46 | Official leaderboard |
| TroubleshootingBench | 44.1% | | #7 of 20 | Lab model card |
| scBench-Long | 21.6% | | #10 of 27 | Official leaderboard |
| DTap Direct Misuse | 28.9% | | #3 of 8 | Official leaderboard |
| DrugDiscoveryBench | 51.6% | | #8 of 20 | Official leaderboard |
| BioTIER Refuse | 75.3% | | #28 of 69 | Official leaderboard |
| ProtocolQA Open-Ended | 37.3% | | #11 of 27 | Lab model card |
| VCT-v2 | 45.3% | | #11 of 25 | Official leaderboard |
| Multimodal Troubleshooting Virology | 53.7% | | #12 of 27 | Lab model card |
| DeepsecBench | 21.0 | | #13 of 29 | Official leaderboard |
| Dictatorship Eval | 82.5% | | #9 of 20 | Official leaderboard |
| BioSecBench-Surveillance | 47.1% | | #13 of 27 | Official leaderboard |
| TxBench-Preclinical-Pharmacology | 55.3% | | #15 of 30 | Official leaderboard |
| DNA sequence design for transcription factor binding | 13.8% | | #2 of 4 | Lab model card |
| Hard-negative protein binding prediction | 1.48% | | #2 of 4 | Lab model card |
| GeneBench | 33.2% | | #1 of 2 | Lab model card |
| SRE-Bench | 3.82% | | #18 of 35 | Official leaderboard |
| CyScenarioBench | 26% | | #9 of 17 | Lab model card |
| ExploitGym | 14.8% | | #10 of 18 | Official leaderboard |
| OncologyBench | 14.7% | | #19 of 30 | Official leaderboard |
| TxBench-Oligo-Discovery | 33.9% | | #19 of 30 | Official leaderboard |
| BioTIER Permit | 99% | | #47 of 69 | Official leaderboard |
| HackBench | 50.9 | | #6 of 8 | Official leaderboard |
| UK AISI cyber ranges | 10% | | #3 of 4 | Lab model card |
| TxBench-Antibody-Discovery | 33% | | #23 of 30 | Official leaderboard |
| VariantBench | 30.5% | | #21 of 27 | Official leaderboard |
| BioSecBench-Function | 28.3% | | #22 of 27 | Official leaderboard |
| First-Person Fairness evaluation | 1.12% | | #10 of 10 | Lab model card |
| CoT-Control | 0.20% | | #9 of 9 | Lab model card |
| Agent JSK | 96.3% | | #7 of 7 | Lab model card |
| OpenAI Deployment Simulation | 0.01% | | #5 of 5 | Lab model card |
| Capture the Flag (Internal) | 88.1% | | #4 of 4 | Lab model card |
| Biochemistry knowledge improvement | 32.3% | | #2 of 2 | Lab model card |
| UK AISI narrow cyber tasks | 90.5% | | #2 of 2 | Lab model card |
| BashArena | 99.1% | | #1 of 1 | Lab model card |