Models

StrategyTune routes AI requests through three frontier models. You pick which one handles a given chat from the model dropdown above the input box. Switching mid-thread is fine — the new model picks up the same conversation.

What's available

These choices reflect what we currently expose; we update them as providers ship new models.

Pricing

Cost is the provider's published rate, no markup. The current per-million-token rates live in the app under Billing & AI → AI Models. They update when the providers do; we don't hardcode them in docs.

For the breakdown of what tokens are billed and how cache pricing works, see AI pricing.

How to choose

• If you don't know, use the default. Claude Opus 4.7 was picked because it's the most reliable on this task: it follows instructions closely, handles the language's strict typing, and recovers well from compile errors.
• Optimize for speed when iterating. Once an indicator is working and you're tweaking colors, parameter ranges, or warmup behaviour, a smaller model is enough.
• Switch providers if one gets stuck. If a model is looping or misreading the spec, switching to a different family often unsticks it.

Caching, briefly

All three providers support prompt caching, which means the system prompt and your chat history are cheaper to replay across consecutive turns of the same conversation. In practice that means:

• The first turn of a fresh chat costs full price.
• Follow-up turns within the same chat are markedly cheaper, because the system prompt and prior turns are read from cache (about 10× cheaper than regular input).
• Anthropic charges a small premium on the cache write (the first turn). OpenAI and Google bill the first turn as plain input.

See AI pricing for the per-token rates and the full cache breakdown.