Cybench: Results as Published

Cybench gives a language-model agent 40 professional capture-the-flag tasks drawn from four distinct competitions chosen to be recent and to span a wide range of difficulty. Each task has its own description and starter files, and it starts in an environment where the agent can run commands and see the output. A task is solved when the answer the agent submits matches the answer key. Subtasks break a task into intermediate steps and are scored separately from fully unguided solutions.

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.