Choosing Your AI Model

CodeSky exposes five providers; each has a sweet spot. The model you pick affects quality, speed, and which plans can use it — this guide helps you choose and switch confidently.

The model matrix

Model	Best for	Latency	Notes
Claude (Sonnet / Opus)	Full apps, complex updates, refactors	Medium	Highest-quality default. Pro+ plans.
Grok Code Fast	Tiny edits, single-file changes	Low	Requires the Grok CLI on PATH.
Gemini 2.5 Pro	Long, structured outputs (CRUD scaffolds)	Medium	Default Free-plan model.
Groq (Llama 3.3 70B)	Quick experiments, very low latency	Very low	API mode — no CLI needed.
Ollama (DeepSeek Coder)	Air-gapped / privacy-first	Depends on hardware	Runs entirely on your machine.

A simple rule of thumb

• Building or refactoring a whole app? Use Claude — it holds the most context and makes the fewest mistakes.
• One quick fix or a tiny tweak? Grok or Groq return in seconds.
• On the Free plan? Gemini is a strong default for scaffolding and CRUD.
• Privacy or offline? Ollama keeps everything on your machine.

Quality vs. speed

• Higher-quality models "think" more, so they're a little slower but need fewer correction rounds — usually a net time saving on real work.
• Fast models are perfect for obvious, well-scoped edits where you already know what you want.
• A common workflow: draft the full build with Claude, then switch to Grok for rapid touch-ups.

Switching mid-project

• Change the model at any point; the next message uses the new provider while the project context is preserved.
• Switching costs nothing — you're not restarting, just changing who does the next edit.

Common mistakes to avoid

• Using a fast model for a big refactor. It'll try, but a higher-quality model gets there in fewer rounds.
• Fighting a "CLI not found" error. That provider's binary isn't installed — just pick Gemini, Groq, or Ollama, which don't need one.

FAQ