skill

skill

PinchBench is a benchmarking system for evaluating LLM models as OpenClaw coding agents.

Agent SkillDevOpen source
Type
Agent Skill
Open source
Yes
GitHub Stars
★ 1.2k
Source
skill-github

Overview

PinchBench is a benchmarking system designed to assess the performance of LLM models as OpenClaw coding agents. It tests model capabilities through real-world tasks such as scheduling meetings, writing code, and handling emails. PinchBench helps developers understand how well models perform in areas like tool usage, multi-step reasoning, handling ambiguous instructions, and generating actual outputs. To integrate with PinchBench, clone the repository and run the script. Ideal for developers looking to evaluate and improve LLM model performance.

Capabilities

  • Tool usage capability
  • Multi-step reasoning capability
  • Handling ambiguous instructions
  • Actual result generation
  • Automated scoring

Use cases

Evaluating LLM models' coding abilitiesComparing performance across different LLM modelsImproving LLM models' real-world application performance

Setup

Requires: Python 3.10+uv 包管理器OpenClaw 实例
git clone https://github.com/pinchbench/skill.git && cd skill && ./scripts/run.sh --model openrouter/anthropic/claude-sonnet-4

This information was compiled by AI from public sources and may contain inaccuracies — please refer to the source.

FAQ

How do I register an API token?

Run the command ./scripts/run.sh --register.

How do I submit official results?

Use the --official-key flag or set the PINCHBENCH_OFFICIAL_KEY environment variable.

Related skills