ClawEval-MM: Results as Published

The multimodal group of Claw-Eval asks an agent to perceive and create, rather than only to call text tools. The paper counts 101 tasks: 53 on video, 22 on documents and images, and 26 on code such as webpage generation, SVG animation, and video editing. Each run is graded for completion, safety, and robustness from execution traces, audit logs, and environment snapshots. A trial passes above 0.75; Pass@3 is the share of tasks passed at least once in three trials, Pass^3 is the share passed every time, and the average score is the mean over the trials.

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.