Choosing a Model

Rolltop connects to every major LLM provider through one app and one credit balance — no juggling separate subscriptions. You choose a model per conversation, and Rolltop keeps the catalog up to date from its servers, so new models appear without an app update.

Providers

Cloud (use credits)

  • Anthropic Claude — the Opus, Sonnet, and Haiku tiers. Strong all-rounders for writing, reasoning, and tool use.
  • OpenAI GPT — the current GPT-5 series, from flagship reasoning down to fast variants.
  • Google Gemini — the Pro and Flash tiers, including million-token context models.
  • xAI Grok — with built-in X Search that can pull cited results from X (Twitter).

Every provider offers roughly the same shape of choice: a frontier model for hard work, a mid model for everyday use, and a cheap fast model for bulk or tool-driven tasks. Pick by the job, not by the vendor.

Local (free, private)

  • Ollama — run any model on your own machine. No credits, no internet required, and your data never leaves your computer. See Local Models (Ollama).

How to pick a model

Choose a model from the selector at the top of a conversation. Rolltop sorts models by recency and highlights the latest releases.

By default the picker shows a curated set of current models — the ones worth reaching for today. Turn on show all models (in Settings → LLM Models) to see the full catalog including older releases. The catalog itself is loaded from Rolltop's servers, so new models appear without an app update, and a superseded model dropping out of the default list never breaks a conversation that still uses it.

model selector dropdown

There are three levels of preference, from broad to specific:

  1. Global preferred model — your default everywhere (set in Settings → LLM Models).
  2. Per-project default — overrides the global default inside a project (also set in Settings → LLM Models, as the Default Model for the active project).
  3. Per-conversation override — pick a different model for one conversation.

Which model should I use?

There's no single right answer, but some rules of thumb:

  • Everyday chat, writing, analysis — a capable general model like Claude Sonnet or GPT.
  • Hardest reasoning / highest quality — a top-tier model like Claude Opus.
  • Fast, cheap, high-volume — a "fast" or "mini/flash" variant.
  • Current events / X discussion — Grok, with its built-in X Search.
  • Private or offline work, zero cost — a local Ollama model.

When in doubt, start with a strong general model and switch if you need more speed or more depth.

Context window

Each conversation can use up to 80% of the model's actual context window — for example around 160K tokens for Sonnet, and up to ~800K for 1M-context models. Longer conversations and large attachments stay within that budget automatically.

Cost

Cloud models consume credits; local models are free. See Credits & Pricing for how usage is charged and where to watch your balance.