---
name: model-chooser
description: Recommend which AI model to build a task with (coding, writing, research, data extraction, documents, support bots, agents, spreadsheets, images, video, transcription, voice, search/RAG, translation), split by one-off versus recurring use. Use when the user asks which model, LLM or AI tool to use, compares models for a job, or is about to pick a model for a build, pipeline or automation.
---

# Model chooser

Recommends the model to build a task with, from GTMagentcy's model chooser (https://gtmagentcy.com/model-chooser/). Recommendations are sourced from vendor docs, public leaderboards and Hugging Face, and carry a checked date.

## When to use

- The user asks "which model should I use for ...", "what's the best AI for ...", or compares models for a job.
- The user is about to choose a model for code, a pipeline, an agent or a one-off piece of work.

## How to answer

1. Work out the frequency before you call anything:
   - **one-off**: done once, by a person, usually in a chat app. Quality first.
   - **recurring**: runs again and again, automated, in production or at volume. Cost, speed and consistency matter.
   If it is genuinely unclear and it would change the answer, ask the user one short question. Otherwise pick the likelier one and say which you assumed.
2. Call the `choose_model` tool from the `model-chooser` MCP server with the user's own description and the frequency.
3. If the matched task looks wrong, call `list_tasks` and then `get_recommendation` with the right `task_id`.
4. When the user wants detail (how to start, code, alternatives, caveats), call `get_recommendation` and use its steps and starter snippet.

## What to say

- Lead with the model and the one-sentence reason.
- Give two alternatives with when to pick each, not the full list.
- Mention the checked date and link the page_url, since models change monthly.
- If the user's constraints rule out the pick (data cannot leave their infrastructure, they are locked to one vendor, budget), recommend the alternative whose "pick it when" matches, and say why.
- Do not invent benchmark numbers or prices. Quote only what the tool returned.

## If the MCP server is not connected

Tell the user they can connect it with:

```
claude mcp add --transport http --scope user model-chooser https://gtmagentcy.com/api/mcp
```

Meanwhile, use the table below and link the matching page: https://gtmagentcy.com/model-chooser/<task_id>-one-off.html or https://gtmagentcy.com/model-chooser/<task_id>-recurring.html. The table is a snapshot from 2026-09-17; the MCP server and the pages are the current source.

| task_id | Task | One-off | Recurring |
|---|---|---|---|
| `coding` | Build or fix software | Claude Opus 5 | Claude Sonnet 5 |
| `writing` | Write copy and long-form content | Claude Opus 5 | Claude Sonnet 5 |
| `research` | Research a market, company or topic | GPT-6 Astra | Claude Opus 5 |
| `extraction` | Classify, tag or extract data | Claude Sonnet 5 | GPT-5.6 Luna |
| `documents` | Read PDFs, contracts and scanned documents | Claude Opus 5 | Mistral OCR 4 |
| `support` | Run a customer-facing chat assistant | Claude Opus 5 | Claude Sonnet 5 |
| `agents` | Automate multi-step work with an agent | Claude Fable 5.1 | Claude Opus 5 |
| `data` | Analyse spreadsheets and numbers | GPT-6 Astra | Gemini 3.8 Flash |
| `image` | Generate or edit images | GPT Image 2.5 | Nano Banana 2 (Gemini 3.1 Flash Image) |
| `video` | Generate video | Gemini Omni Flash | MiniMax H3 |
| `transcription` | Transcribe audio, calls and meetings | ElevenLabs Scribe v2 | ElevenLabs Scribe v2 |
| `voice` | Generate voiceover and speech | ElevenLabs v3 | Cartesia Sonic 3.6 |
| `search` | Search and answer from your own content | Gemini 3.8 Flash | Gemini Embedding 2 |
| `translation` | Translate and localise | Claude Opus 5 | Gemini 3.8 Flash |