BlogLLM Benchmark Scores as Published, Side by Side
WMDP: Results as Published
Every result the catalog holds for WMDP: 3 models from 1 developer, read from 1 page. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area BiosecurityMaintainer Center for AI SafetyIntroduced 2024Data version 2026103001, as of 2026-10-30
WMDP is a set of multiple-choice questions that serve as a proxy for hazardous knowledge in biosecurity, cybersecurity and chemical security. The questions were filtered by experts to remove sensitive and export-controlled details. Models are scored by accuracy, and the same questions are used in unlearning research, where the aim is to lower accuracy on the hazard questions while keeping general ability.
WMDP-Bio
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Grok 4.7xAI, released 2026-09-21 | 88.1% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Grok 4.6xAI, released 2026-08-12 | 90% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Grok 4.5xAI, released 2026-07-16 | 90.9% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
WMDP-Chem
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Grok 4.7xAI, released 2026-09-21 | 84.9% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Grok 4.6xAI, released 2026-08-12 | 85.3% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Grok 4.5xAI, released 2026-07-16 | 87.3% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
WMDP-Cyber
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Grok 4.7xAI, released 2026-09-21 | 88.1% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Grok 4.6xAI, released 2026-08-12 | 90.1% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
| Grok 4.5xAI, released 2026-07-16 | 83.2% | xAI, Grok 4.7 model cardchecked 2026-10-14 | Effort: high |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.