# Models & Cost Reference

> **Summary — what this page covers**
> A keeper page: the current model lineup with per-token pricing, the cost-reduction levers, and
> a clear map of what genuinely needs paid access versus where a free/local alternative works.
> Verify the rates against current pricing before the event.

***Free-first: default to free/local where possible; pay only where it's genuinely required.***

## Model lineup & pricing

Per million tokens, input / output. *(Verify against current pricing before the event.)*

| Model | Input / Output (per 1M tokens) | Use for |
|---|---|---|
| Claude Haiku 4.5 | $1 / $5 | High-volume, mechanical work; CI code review |
| Claude Sonnet 4.6 | $3 / $15 | Default for app features |
| Claude Opus 4.8 | $5 / $25 | Hardest reasoning |

## Cost-reduction levers

- **Prompt caching** — up to ~90% off the cached input portion (Section 1).
- **Batch processing** — ~50% cheaper for non-interactive workloads.
- **Model choice** — Haiku for CI reviews and mechanical tasks (Section 5).

## Paid vs free/local map

| Capability | Needs paid? | Free / local alternative |
|---|---|---|
| Claude API calls (all sections) | **Yes** — API key required | Trial credit covers the workshop |
| RAG embeddings (Section 3) | OpenAI key by default | **Ollama + `nomic-embed-text` (768-dim)** behind the same interface |
| Vector store (Section 3) | No | **Qdrant** in Docker (local) |
| Integration test infra (Section 5) | No | **TestContainers** (local Docker) |
| Claude Code (Day 1) | Pro min ($20/mo) | — (not on Free) |

> **Remember:** a Claude.ai **Pro/Max subscription does NOT grant API access.** Day 2 runs on the
> Anthropic **API**, which is billed separately (pay-as-you-go) — you need a **Console key**.