Clusydocs
Models & tools

Choosing a model

Choose a model by task, plan, and reasoning effort.

Choose the model from the picker under the chat input. The choice is per message, so you can use a fast model for routine work and switch up only for a hard step.

Built-in models

ModelGood for
AutoThe default for routine transforms, inspection, and straightforward plots, at a fixed Medium effort.
DeepSeek V4 FlashFast, low-cost everyday agent work, with adjustable effort.
DeepSeek V4 ProMore involved reasoning and multi-step workflows at a moderate cost.
Kimi K2.7 CodeCode-heavy agent work and a large context window.
GLM 5.2Agentic coding and reasoning with a large context window.
Qwen3.8 MaxDeep, always-on reasoning and very long context.
Kimi K3Long-context, deep agentic reasoning.
Sonnet 5Balanced frontier reasoning for serious analysis and modeling.
Fable 5.1Anthropic's most capable model, for the longest and hardest agentic work, at the highest cost band.
Opus 5The latest Opus flagship for the hardest debugging, design, and multi-step work.
Opus 4.8The prior Opus flagship and a proven agentic workhorse.
GPT-5.6 SolOpenAI's built-in frontier model for difficult work or a second opinion.

Kimi K2.6 and Opus 4.7 remain selectable under Show older models for compatibility. Kimi K2.7 Code now has one standard serving option; the former Fast variant has been retired, and an older saved selection resolves to the standard model.

The $ marks in the picker are relative cost bands, not exact prices. The current allowance impact appears in Settings → Billing.

Which plan reaches which model

PlanBuilt-in models
FreeAuto only.
PlusAuto, DeepSeek V4 Flash, and DeepSeek V4 Pro.
ProEverything on Plus, plus Kimi K2.6, Kimi K2.7 Code, Kimi K3, GLM 5.2, and Qwen3.8 Max.
Max 10x / 30xEverything on Pro, plus Sonnet 5, Opus 5, Opus 4.8, Opus 4.7, Fable 5.1, and GPT-5.6 Sol.

Max 10x and Max 30x have the same built-in model catalog; their usage, GPU, RAM, storage, and sandbox limits differ. See Plans.

Models outside the plan table

These routes work on every plan, including Free, and their model inference does not use the Clusy allowance:

  • Bring your own key: an Anthropic key unlocks Sonnet 5, Opus 4.8, and Fable 5.1; an OpenAI key unlocks GPT-5.6 Sol. Your provider bills the calls.
  • Local runner: a paired machine can expose Claude Code Sonnet, Claude Code Opus, or Codex through the CLI subscription on that machine.

BYOK and local-runner rows appear only while the corresponding integration or runner is available.

How to choose

The task is…Start with
Mechanical, like a reformat, rename, or simple plotAuto or DeepSeek V4 Flash
Normal analysis, coding, and modelingDeepSeek V4 Pro, Kimi K2.7 Code, or GLM 5.2
Long-context or difficult open-model workQwen3.8 Max or Kimi K3
High-stakes design, subtle debugging, or a costly decisionSonnet, Opus, Fable 5.1, or GPT-5.6 Sol

The model is one dial and reasoning effort is the other. Auto has one fixed Medium setting; other models expose only the effort levels they actually support.

Defaults and custom instructions

Set the model and effort used by new sessions in Settings → Models. Per-message choices still override them. The same section holds custom instructions—up to 2,000 characters of standing guidance such as your preferred libraries, naming conventions, or plot style.

On this page

Ask docs