agentscope-model-compare

该 Skill 运行一个 Python 脚本，把同一 AgentScope 任务分别提交给多个 LLM 模型（OpenAI 及可选的阿里云 DashScope），统计延迟、Token 消耗与输出长度，并打印各模型完成或失败的对比表。需要配置相应 API 密钥，调用会产生费用，任务文本会发送至外部模型服务。

trend0522

热度 7.65

最近更新 2026-10-03 09:44:52

分类：数据依赖：代码执行 / 第三方 API / 内部调用大模型 / 读写文件应用场景：报告撰写 / 研究

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add agentscope-model-compare

Metadata

name: agentscope-model-compare
description: Compare multiple LLM models for use with AgentScope 2.0. Use this whenever the user wants to evaluate models, benchmark responses, or choose the best model for their agent task.
version: 0.1.0
metadata: {"openclaw":{"requires":{"env":["OPENAI_API_KEY","DASHSCOPE_API_KEY"],"bins":["python3"]},"primaryEnv":"OPENAI_API_KEY","envVars":[{"name":"OPENAI_API_KEY","required":true,"description":"OpenAI API key for OpenAI models."},{"name":"DASHSCOPE_API_KEY","required":false,"description":"Optional DashScope API key for Alibaba Cloud models."}],"emoji":"⚔️","homepage":"https://github.com/agentscope-ai/agentscope"}}

AgentScope Model Compare

This skill runs the same AgentScope task against multiple LLM models and compares latency, token usage, and output style.

What this skill does

Configures multiple AgentScope chat models
Runs a shared agent task against each model
Reports latency, token counts, and output length
Highlights which models completed vs failed

Usage

Run the comparison script from this skill folder:

python3 scripts/compare_models.py --task "Translate this into Chinese: Hello world"

The script uses environment variables for API keys and prints a comparison table.

Safety

API keys are read from .env, which is gitignored.
No keys are hardcoded.
Costs may apply; limit concurrent calls if needed.