A data-driven intelligence hub for tracking AI model capabilities and costs to ensure optimal compute routing.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install model-benchmarks
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install model-benchmarks using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Model Benchmarks?
Model Benchmarks is a comprehensive intelligence tool designed for the Openclaw Skills ecosystem to eliminate guesswork in model selection. By aggregating real-time data from premier evaluation platforms like LMSYS Chatbot Arena, HuggingFace, and BigCode, this skill provides developers with a standardized 0-100 scoring system across various domains including coding, reasoning, and creative writing.
The primary purpose of this addition to Openclaw Skills is to bridge the gap between performance and cost. It calculates performance-per-dollar ratios, allowing users to identify high-efficiency models (like Gemini 2.0 Flash) that can replace expensive flagship models for specific tasks without sacrificing quality. This ensures that your AI agents are always using the most efficient compute resources available.
Model Benchmarks Use Cases
- Optimizing API spend by routing simple queries to low-cost, high-performance models.
- Selecting the best-in-class model for specific technical tasks like complex Python programming or mathematical reasoning.
- Tracking the performance decay or improvement of specific LLM versions over time.
- Automating model fallbacks within Openclaw Skills based on real-time capability scores.
- Generating cost-efficiency reports for enterprise AI deployments.
How Model Benchmarks Works
- The skill fetches the latest capability data and rankings from multiple global leaderboards and benchmarking platforms.
- It normalizes diverse metrics into a unified 0-100 capability score for different task categories.
- The system calculates cost-efficiency ratios by comparing current API pricing against performance scores.
- Users query the skill via the CLI to receive ranked recommendations based on their specific task requirements (e.g., coding vs. translation).
- The output can be integrated directly into agent configuration files to dynamically update model routing logic.
Model Benchmarks Setup
To integrate this capability into your Openclaw Skills environment, use the following commands:
# Fetch the latest model intelligence data
python3 skills/model-benchmarks/scripts/run.py fetch
# Get a recommendation for a specific task type
python3 skills/model-benchmarks/scripts/run.py recommend --task coding
# Query specific model details
python3 skills/model-benchmarks/scripts/run.py query --model gpt-4o
Model Benchmarks Data Schema & Taxonomy
The skill organizes model intelligence using a multi-dimensional taxonomy:
| Platform | Update Frequency | Capabilities Tracked |
|---|---|---|
| LMSYS Arena | Daily | General, Reasoning, Creative |
| BigCode | Weekly | HumanEval, MBPP Coding |
| Open LLM | Daily | Knowledge, Comprehension |
| Alpaca Eval | Weekly | Instruction Following |
Internal task mapping routes coding tasks to models with high Reasoning + Coding scores, while writing tasks prioritize Creative + General scores.
Model Benchmarks Advanced Features
- Custom Benchmark Sources: Ability to modify the source scripts to include internal or private benchmarking data.
- Task-Specific Weighting: Customize the importance of different capabilities (e.g., prioritize speed over reasoning) within the scoring algorithm.
- Automated Routing Integration: Scripted hooks to update Openclaw Skills configuration files automatically based on the latest fetch results.
- Trend Analysis: Compare model performance changes over a 30-day window to identify model drift or improvements.
- Enterprise Export: Generate CSV or JSON reports for cost monitoring and stakeholder dashboards.
SKILL.md
Loading
Related Openclaw Skills
Stratos Storage
A specialized skill for seamless file interaction with the Stratos Decentralized Storage (SDS) network.
Obul Proxy
Automates x402 payment negotiation for AI agents by proxying upstream requests through Obul.
Grok Search
Perform high-speed web and X (Twitter) searches with structured JSON output and verified citations using the xAI Grok API.
Presearch Search
A privacy-focused, decentralized search skill that provides uncensored web results for AI agents via a distributed node network.
Argos Product Research Skill
An expert AI product researcher for Argos.co.uk that enables natural language search, side-by-side comparisons, and sentiment-based review analysis.
Pomodoro Bot
An automated Pomodoro assistant that manages structured 25-minute work blocks and 5-minute breaks to optimize deep focus.
Featured*
