BlogLLM Benchmark Scores as Published, Side by Side
CVE Rediscovery (Anthropic internal): Results as Published
Every result the catalog holds for CVE Rediscovery (Anthropic internal): 3 models from 1 developer, read from 2 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area CybersecurityMaintainer Anthropic (internal evaluation)Introduced 2026Data version 2026111201, as of 2026-11-12
The model gets a snapshot of a real open-source repository at a known-vulnerable version, without its commit history, and the list of files the fix changed, with no internet access, and has one attempt to find security bugs. A model judge compares its reports with the published advisory and fix. A task counts as solved when any report has the same root cause; the set is 1,829 advisories published in October 2026.
CVE Rediscovery
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Claude Mythos 5.2Anthropic, released 2026-11-09 | 89.6% | Anthropic, Claude Fable 5.2 and Mythos 5.2checked 2026-11-09; printed on 1 more page | - |
| Claude Opus 5.5Anthropic, released 2026-09-22 | 81.3% | Anthropic, Claude Fable 5.2 and Mythos 5.2checked 2026-11-09; printed on 1 more page | - |
| Claude Mythos 5.1Anthropic, released 2026-09-01 | 78.3% | Anthropic, Claude Fable 5.2 and Mythos 5.2checked 2026-11-09; printed on 1 more page | - |
Every number here is read from the same catalog file as the benchmark table, data version 2026111201, 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.