# AI Model Discovery Protocol

URL: https://research-memex.org/docs/craft/ai-model-discovery-protocol
Description: Model choice improves when you test candidates against your own research tasks.

*The right model is the one that fits your materials, not the leaderboard.*

<Aside>
Read this if you have several frontier or specialist models and want a defensible way to test which ones fit your research.
</Aside>

The [AI Model Reference Guide](/docs/craft/ai-model-reference-guide) tells you what is available. This page is the protocol for finding what fits you: a structured experiment against your own research materials.

## Getting started [#getting-started]

<Steps>
  <Step title="Phase 1: capability discovery" id="phase-1-discovery">
    Run a one- or two-day sampling pass.

    1. Choose one real task: theory synthesis, methods critique, extraction, or writing revision.
    2. Choose candidates from the [current model reference](/docs/craft/ai-model-reference-guide#current-model-families). Compare at least 3 families you can access, including a lower-cost candidate. Use the same prompt and evidence pack for each.
    3. Record differences in style, depth, accuracy, and failure modes.
    4. Note the access path: web app, API key, Cherry Studio, Claude Code, Codex, Antigravity CLI, OpenCode, or Vox MCP.
  </Step>

  <Step title="Phase 2: control surface testing" id="phase-2-controls">
    Do not assume temperature is available or useful.

    1. Start with provider defaults.
    2. Check the exact model and access path in the [model settings reference](/docs/craft/ai-model-reference-guide#configuration-defaults-first) and the provider's linked documentation. Test supported effort or thinking settings. Some models always reason. Do not assume you can turn reasoning off or carry settings across versions.
    3. Use Sequential Thinking MCP when you want visible, revisable steps across models.
    4. Only test temperature on models whose current docs support it.
  </Step>

  <Step title="Phase 3: task matching" id="phase-3-task-matching">
    Build a small table for your own work:

    | Task | Best model | Backup | Why |
    | --- | --- | --- | --- |
    | Screening | | | |
    | Extraction | | | |
    | Theory synthesis | | | |
    | Methods critique | | | |
    | Draft revision | | | |
    | Code/data analysis | | | |

    The "why" column matters. If you cannot explain why a model won, keep testing.
  </Step>

  <Step title="Phase 4: personal strategy" id="phase-4-strategy">
    Decide the operating pattern you will actually use:

    - primary model: the model you trust for high-stakes reasoning.
    - volume model: the model cheap enough for iteration.
    - challenge model: the model that sees different weaknesses.
    - fallback model: the model you can work with when quota, latency, or outages interrupt the work.
    - privacy path: the local or constrained workflow for sensitive materials.
  </Step>
</Steps>

## Evaluation notes [#evaluation-notes]

For every test, save:

- the exact prompt
- the model and access path
- the evidence pack or files used
- the control settings, especially reasoning effort or thinking level
- the failure mode, not just the useful output

Start with provider defaults, test reasoning controls next, and adjust sampling only where the provider supports it. Once you know which models fit your work, choose an access route in your workspace: [Cherry Studio](/docs/implementation/agentic-ai-tools/cherry-studio-setup-guide), [Claude Code](/docs/implementation/agentic-ai-tools/claude-code-setup-guide), [Codex](/docs/implementation/agentic-ai-tools/codex-setup-guide), [Antigravity CLI](/docs/implementation/agentic-ai-tools/antigravity-cli-setup-guide), [OpenCode](/docs/implementation/agentic-ai-tools/opencode-setup-guide), or [Vox MCP](/docs/toolkit/vox-mcp).