Claw Eval Agent Benchmark with 24 Task Categories

139 agent tasks across general and multimodal splits evaluate real-world AI agent performance. The benchmark covers 24 categories including communication, finance, and operations, created by claw-eval. It was last updated in March 2026.

Use Cases

• Benchmark agent performance across the 24 task categories like communication and finance
• Evaluate multimodal agent capabilities on tasks requiring perception and creation, such as webpage generation
• Analyze task_id results to compare agent strategies on the general and multimodal splits

Strengths

• 139 total examples across two defined splits
• Covers 24 distinct real-world task categories
• Includes a dedicated multimodal split of 35 tasks

Limitations

• Small scale with only 139 total task examples
• Unknown data structure, columns, and sample data details limit reproducibility
• Potential bias from unknown data collection and curation methods

Provenance

Source

claw-eval on Hugging Face

Collection Method

Benchmark dataset for evaluating AI agents

Time Range

null

Freshness

Last updated March 2026

Geography

null