Benchmark OpenClaw coding agents against repeatable real tasks before rollout with PinchBench

Run a real-task benchmark suite against OpenClaw agents so model or harness changes can be compared before they hit production workflows.

Prerequisites

Running OpenClaw instance, Python 3.10+, uv, PinchBench repository checkout, model provider credentials as documented upstream

Installation

Use the upstream install or setup path that matches your environment:

• git clone https://github.com/pinchbench/skill.git

Requirements and caveats from upstream:

• Note: Model IDs must include their provider prefix (e.g. openrouter/, anthropic/). OpenRouter is the default provider used for routing.
• Python 3.10+

Basic usage or getting-started notes:

• Tool usage — Can the model call the right tools with the right parameters?

• bash

• Clone the skill

• Source: https://github.com/pinchbench/skill

• Extracted from upstream docs: https://raw.githubusercontent.com/pinchbench/skill/HEAD/README.md

Documentation

• https://pinchbench.com

Source

• Agent Skill Exchange