plugin.json

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "whizi-model-finder",
  "version": "1.0.0",
  "description": "Recommends and compares AI models (LLMs) for a task, with list prices, context windows and measured scores in 30 languages. Read-only, no account needed.",
  "author": {
    "name": "Whizi",
    "email": "whizihelp@gmail.com",
    "url": "https://whizi.io"
  },
  "homepage": "https://whizi.io/tools/model-cost-index",
  "license": "UNLICENSED",
  "keywords": [
    "ai models",
    "llm",
    "model comparison",
    "pricing",
    "benchmarks",
    "multilingual"
  ],
  "extensions": {
    "com.openai": {
      "interface": {
        "displayName": "Whizi Model Finder",
        "shortDescription": "Pick the right AI model",
        "longDescription": "Whizi Model Finder tells you which AI model to use for a job, and how named models compare.\n\nAsk which model is best for coding, writing, long documents, screenshots, math, data analysis, image generation or high-volume work, and it returns three picks: best quality, best value and lowest cost. Each pick shows the list price per million tokens, the context window and a one-line reason drawn from public leaderboards and the providers' own launch numbers.\n\nAsk about a language and the picks come from the Whizi Multilingual LLM Bench, which measures models in 30 languages including Arabic, Hindi, Japanese, Persian and Vietnamese.\n\nName two to four models and it compares them side by side: price, context window, image input, multilingual score and the tasks Whizi recommends each for.\n\nIt is read-only. It needs no account, sends nothing to an AI provider and stores nothing you ask. Every card links to the full model page on whizi.io.",
        "developerName": "Whizi",
        "category": "Productivity",
        "capabilities": [
          "Read"
        ],
        "websiteURL": "https://whizi.io",
        "supportURL": "https://whizi.io/support",
        "privacyPolicyURL": "https://whizi.io/privacypolicy",
        "termsOfServiceURL": "https://whizi.io/termsofservice",
        "defaultPrompt": [
          "Which AI model is best for coding, and which is cheapest?",
          "What is the best AI model for chatting in Arabic?",
          "Compare Claude Opus 5.5 and GPT-6 Sol."
        ],
        "brandColor": "#2F6FE4",
        "brandColorDark": "#7AA7FF",
        "composerIcon": "./assets/composer-icon.png",
        "logo": "./assets/logo.png",
        "screenshots": [
          "./assets/screenshot-1-recommend.png",
          "./assets/screenshot-2-language.png",
          "./assets/screenshot-3-compare.png"
        ]
      },
      "publication": {
        "release_notes": "First release. Two read-only tools: recommend_model (three picks for a task or language) and compare_models (two to four named models side by side), each rendered as a card."
      },
      "review": {
        "demo_recording_url": "https://whizi.io/plugin/model-finder-demo.mp4",
        "test_cases": {
          "positive": [
            {
              "description": "Recommend models for a task",
              "prompt": "Which AI model is best for coding?",
              "tools_triggered": "recommend_model",
              "expected_behavior": "Calls recommend_model with task \"coding\". A card shows three picks labelled Best quality, Best value and Lowest cost, each with provider, price per million tokens, context window and a one-line reason. With data as of 2026-10-09 these are Claude Opus 5.5, Claude Sonnet 5.5 and DeepSeek V4.1 Flash. The reply adds reasoning without repeating the whole table."
            },
            {
              "description": "Recommend models for a specific language",
              "prompt": "What's the best AI model for chatting in Arabic?",
              "tools_triggered": "recommend_model",
              "expected_behavior": "Calls recommend_model with task \"multilingual\" and language \"ar\". The card ranks picks by their measured Arabic score on the Whizi Multilingual LLM Bench and shows that score on each pick (80 for the top two as of 2026-10-09)."
            },
            {
              "description": "Cost-first recommendation for long documents",
              "prompt": "I need to summarize a 300-page contract as cheaply as possible. Which model should I use?",
              "tools_triggered": "recommend_model",
              "expected_behavior": "Calls recommend_model with task \"long_documents\" and priority \"cost\". The Lowest cost pick is listed first, and every pick has a context window of at least 1M tokens."
            },
            {
              "description": "Compare two named models",
              "prompt": "Compare Claude Opus 5.5 and GPT-6 Sol. Which should I choose?",
              "tools_triggered": "compare_models",
              "expected_behavior": "Calls compare_models with both model names. The card shows both models side by side with price per million tokens, context window, image input, multilingual score and the tasks Whizi picks each for. The reply weighs the trade-off from the card's numbers: Opus 5.5 is Whizi's pick for eight task types, while GPT-6 Sol costs half as much per token and scores higher on the multilingual bench (63.3 against 60.7 as of 2026-10-09)."
            },
            {
              "description": "Recommend models for image input",
              "prompt": "Which AI model is best at reading screenshots and charts?",
              "tools_triggered": "recommend_model",
              "expected_behavior": "Calls recommend_model with task \"vision\". The card shows three picks that all accept image input, each with price, context window and a reason citing a vision benchmark."
            }
          ],
          "negative": [
            {
              "description": "Doing the task itself, not choosing a model. The plugin only recommends models, so ChatGPT should answer directly.",
              "prompt": "Write a Python function that reverses a string."
            },
            {
              "description": "Unrelated question. Nothing about AI models, so the plugin should not be called.",
              "prompt": "What's a good recipe for banana bread?"
            },
            {
              "description": "Asks for an image to be generated. The plugin cannot run models or create images, so ChatGPT should handle it without calling the plugin.",
              "prompt": "Generate a picture of a cat wearing a space helmet."
            }
          ]
        }
      },
      "apps": "./.app.json"
    }
  }
}