Features
Two-stage scoring: HTTP interception + LLM judge
283 live tasks across 163 live platforms
Public leaderboard with cost/task metrics
Multiple evaluation harnesses (Claw-Eval, WildClawBench, ClawMark)
Open-source dataset and traces on Hugging Face
CLI tool: pip install clawbench-eval
Supports V1 and V2 corpora
Scoring rubrics: lenient and strict reward
Gradio Space for interactive leaderboard browsing
Model submission and comparison support
Task gallery and trace visualization
Integration with other benchmarks (WebVoyager, OSWorld)
5-layer trace bundles with video, HAR, agent reasoning
JSONL-native dataset for SFT/DPO/PRM-ready fine-tuning
Weekly refreshed traces with 283+ tasks
Undetectable AI essay generation
Access to 12,000+ academic source providers
10M+ real academic sources with free PDFs
Automatic reference lists (MLA, APA, Chicago, etc.)
Grammar and spelling checker
AI humanizer to bypass GPTZero, Originality.ai
Built-in originality checker
Outline generator for structured writing
Literature review assistance
Case study support with real-world data
Citation generator for multiple styles
AI suggestion for arguments and refinements
Multi-language support (30+ languages)
