WildClawBench: Results as Published

WildClawBench is a set of 60 human-authored tasks for agents that already run in a command-line harness. The tasks are bilingual and multimodal and fall into six categories. Each task runs in a Docker container with a real harness, such as OpenClaw, Claude Code, Codex, or Hermes Agent, and with real tools rather than mock services. Grading mixes fixed rule checks, an audit of side effects in the environment, and a language or vision-language judge.

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.