# pinchbench/skill

PinchBench is a benchmarking system for evaluating LLM models as OpenClaw coding agents. Made with 🦀 by the humans at https://kilo.ai

Repository: https://github.com/pinchbench/skill
Canonical: https://ross.abutalabs.com/products/pinchbench-skill
Homepage: https://pinchbench.com
Language: Python
License: MIT
License Family: permissive
Last push: 2026-07-02T01:44:55+00:00

## Health v2 (maintenance only)
Score: 72/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 90, release rhythm 83, longevity 14
- inputs: {"age_days": 204, "days_push": 63, "days_rel": 119, "gap_med": 0.0, "n_releases_24m": 13}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1325, forks 154 (observed 2026-08-28T04:04:22.305116+00:00)

## What it is
PinchBench is a benchmarking system that evaluates LLM models as OpenClaw coding agents using 53 real-world tasks like scheduling, coding, research, and email triage. This repository contains the benchmark skill/task definitions, with results published on a public leaderboard at pinchbench.com.

## Use cases
- compare LLM models for coding agent performance
- evaluate which model is best for my AI agent
- benchmark LLM tool usage and multi-step reasoning
- find the best value or cheapest model for agent tasks
- run real-world agent tasks against different models
- measure LLM success rate, speed, and cost
- add custom benchmark tasks for AI agents

## When to choose
- You run OpenClaw (or similar) agents and need to pick the best model
- You want real-world task benchmarks rather than synthetic LLM tests
- You need cost/speed/success-rate comparisons across many models via OpenRouter

## When to avoid
- You need standard academic benchmarks like SWE-bench or HumanEval
- You don't use OpenClaw or an equivalent agent runtime
- You want to benchmark non-agent capabilities like raw code completion

## Facets
- artifact type: dataset
- maturity: active
- function: benchmarking, llm-inference, agent-framework, testing, data-generation
- domain: large-language-models, machine-learning, developer-tools, testing
- platform: python, cli, cross-platform
- tags: llm-benchmark, coding-agents, leaderboard, openclaw, task-suite, llm-judge, evaluation, ai-agents

## Member repositories
- pinchbench/skill (main) score 72

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:22.305116+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:46:35.513444+00:00, confidence not recorded.
  - readme: https://github.com/pinchbench/skill (fetched 2026-08-28T04:04:22.305116+00:00, sha b5c2e6c273e1)
  - homepage: https://pinchbench.com (fetched 2026-08-29T12:05:32.062502+00:00, sha 1e488dfa9617)
  - site_page: https://pinchbench.com/about (fetched 2026-08-29T12:05:32.065026+00:00, sha ea2d53367468)
- Data as of 2026-08-30T08:39:29.467469+00:00.