Primary navigation

Models

The AI models available in ZeroTwo Work and ZeroTwo, and what each is for

ZeroTwo is a multi-vendor router, not a single-model product. The catalog spans ZeroTwo, Anthropic, Google, xAI, Meta, Qwen, DeepSeek and others, and you pick the model per chat.

Auto is ZeroTwo's router. It resolves to GPT 6 Luna for chat and agent runs, and paid plans include a monthly allowance of free Auto requests.

The ZeroTwo GPT line in the catalog is the usual starting point for agent work.

  • GPT 6.1 Sol: a premium reasoning model for complex coding, research and multi-step analysis, with a 1.05M-token context window.
  • GPT 6 Sol: the GPT 6 flagship, with the same window.
  • GPT 6 Luna: a lower-cost GPT 6 model, and the first entry on the Free plan's pinned list.
View other models

GPT 5.6 Sol, Terra and Luna and GPT 5.5 remain in the catalog alongside the GPT 6 line.

Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5, plus Gemini, Grok, Qwen and other families, appear in the same switcher. Which models you can use without spending credits depends on your plan; see pricing.

Free plan models

The Free roster is pinned at the top of the switcher in this order: GPT 6 Luna, Muse Spark 1.3 (c), GLM 5.3 Flash, MiniMax M3 and DeepSeek V4.1 Flash. Claude Haiku 5.5, Gemini 2.5 Flash Lite, Muse Spark 1.2 (c) and MiniMax M3.1 Flash follow. Other models show a locked or credit chip. If your saved model needs a paid plan or purchased credits you do not have, ZeroTwo switches to an included model when you send and tells you which one.

Choosing a model

Match the model to how hard the job is and how much you want to spend. Start with Auto or a Luna-class model for routine questions, and move to a Sol-class or other flagship model for long, multi-step work that needs more reasoning. Compare the credit chips before you commit a large task.

Pick a reasoning effort

Each model declares its own reasoning ladder, so the available settings differ.

  • In plain Chat, the composer dropdown offers None, Standard and Extended, mapped to the provider's native settings.
  • In Work and Code, you see the model's real rungs. GPT 6.1 Sol offers Low, Medium, High, X-High and Max with Medium as the default. GPT 6 Sol and GPT 6 Luna add None. Claude Opus 5.5 offers Off through Max, Claude Haiku 5.5 offers Low through Max, and Claude Sonnet 5.5 stops at High.
  • /reasoning sets the level from the slash menu in Work and Code. On desktop, shortcuts to cycle, raise or lower the effort exist but have no default key, so assign them under Settings > Shortcuts.

Raise effort for planning-heavy tasks and lower it for quick ones. Because ladders differ per model, tune from a task you know well.

Fast mode

The desktop shortcut Toggle Fast mode (unbound by default) drops the composer to its lowest reasoning rung and back. It is a reasoning shortcut, not a separate faster service tier.

Route subagent work to other models

Under Settings > General > Advanced, turn on Sub-agent Model Routing to let the agent send hard, medium and easy tasks to models you configure. See subagents for how delegated runs work.

Use your own subscription or key

On the desktop app, some model rows run on accounts you already hold and bill to those accounts instead of ZeroTwo credits. Examples are ZeroTwo (Your Subscription), Claude (Your Subscription), Grok Build (Your Subscription), MiniMax (Your Subscription), OpenRouter (Your Key) and Kimi Code (Your Key). Connect them under Settings > Providers.