Claw-Eval: Results as Published

Claw-Eval is an end-to-end suite for agents doing multi-step software work. It has 300 human-checked tasks in nine categories and three groups: general workflows (161), multimodal perception and creation (101), and multi-turn professional dialogue (38). Graders use execution traces, audit logs, and environment snapshots to score completion, safety, and robustness. Over three trials, the average score is the mean, Pass@3 is the share of tasks that pass at least once, with a pass defined as a score above 0.75, and Pass^3 is the share that pass every time.

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.