# Module 2 — Choosing the Right Model

> Models are not interchangeable. Each has a personality, a strength, and a cost profile. Using the wrong model is like using a sledgehammer to hang a picture.

---

## The Models

### GPT-4.1 (OpenAI)
**Personality:** Fast, practical, broadly capable. The generalist.

**Strengths:**
- Instruction-following — does what you ask, precisely
- Code generation across all major languages
- Tool use and function calling
- Strong at structured output (JSON, tables, forms)

**When to use:**
- Everyday coding tasks and quick answers
- Generating boilerplate or templates
- When speed matters more than depth
- Integrations that require OpenAI's API ecosystem

**Context window:** 1M tokens  
**Best via:** ChatGPT, Cursor, API

---

### Claude Opus 4 (Anthropic)
**Personality:** Deep, careful, honest. The thinker.

**Strengths:**
- Long-context reasoning — holds more in its head at once
- Nuanced writing and editing
- Complex multi-step analysis
- Refuses less, reasons more transparently

**When to use:**
- Long documents, research synthesis, strategy work
- When you need the model to *think*, not just respond
- Legal, financial, or compliance-sensitive drafting
- Anything where accuracy matters more than speed

**Context window:** 200K tokens  
**Best via:** Claude.ai, API, Claude Code

---

### Claude Sonnet 4.5 (Anthropic)
**Personality:** Balanced. Fast enough, smart enough, affordable.

**Strengths:**
- Best price-to-capability ratio in the Anthropic lineup
- Excellent at agentic tasks (multi-step, tool use)
- Strong coding + reasoning without Opus's cost

**When to use:**
- Production workloads and automated pipelines
- Agentic tasks where the model needs to plan and execute
- When you want Opus-level thinking at half the cost

**Context window:** 200K tokens  
**Best via:** API, OpenClaw, Claude Code

---

### Gemini 2.5 Flash (Google)
**Personality:** Ultra-fast, cheap, multimodal. The sprinter.

**Strengths:**
- Fastest time-to-first-token of any frontier model
- Native image, audio, and video understanding
- Huge context window (1M tokens)
- Very low cost — ideal for high-volume tasks

**When to use:**
- High-volume batch processing
- Image/video/audio analysis at scale
- Real-time applications where latency is critical
- When you're running thousands of calls and cost matters

**Context window:** 1M tokens  
**Best via:** Google AI Studio, API, Vertex AI

---

## Model Selection Guide

| Task | Best Model |
|------|-----------|
| Quick code fix | GPT-4.1 |
| Deep research synthesis | Claude Opus 4 |
| Production API pipeline | Claude Sonnet 4.5 |
| Image/video analysis | Gemini 2.5 Flash |
| Multi-step agent task | Claude Sonnet 4.5 |
| Long document review | Claude Opus 4 |
| High-volume batch jobs | Gemini 2.5 Flash |
| Complex strategy work | Claude Opus 4 |
| Fast prototyping | GPT-4.1 or Sonnet |

---

## Cost Awareness

Running AI at scale costs real money. Rules of thumb:

1. **Don't use Opus for tasks Sonnet can handle.** You'll pay 5x for no gain.
2. **Batch jobs belong on Flash or Sonnet.** Never Opus.
3. **Cache your prompts.** Most providers charge less for cached tokens.
4. **Measure before optimizing.** Log your token usage for one week before cutting costs.

---

## The Mental Model

Think of models like staff, not software:

- **GPT-4.1** = the reliable generalist who executes quickly
- **Opus** = the senior analyst you bring in for hard problems  
- **Sonnet** = the workhorse who runs your operations
- **Flash** = the intern who processes 1,000 things overnight

---

*Next: [Module 3 — The Executor Role](executor.md)*