Skip to main content

Overview

CREAO gives you access to multiple AI models from different providers. Choose the model that best fits your task — whether you need maximum intelligence, a large context window, or cost efficiency. Select a model from the model dropdown at the top of the chat interface. Your choice persists per thread, so you can use different models for different conversations.
Free-plan users see a simplified tier picker instead, currently set to Lite — CREAO automatically selects a good model for that tier. Upgrading to a paid plan unlocks Standard and Advanced tiers plus hand-picking a specific model.

Model comparison

All models have full access to the same tools — code execution, web search, image generation, file handling, and all connected skills and integrations.
The list is grouped by provider, and each group leads with that provider’s current models. Superseded releases collapse under an Older versions row inside their own group — see How the list is ordered.

Current models

Older versions

Each of these is superseded by a newer model from the same provider. They stay fully selectable and identically priced — they simply sit one click away, under their provider’s Older versions row.

How the list is ordered

The model dropdown follows the same four rules everywhere it appears — chat, agent default model, and the tier picker:
Models are grouped under a provider heading, in a fixed order: Anthropic, Google, OpenAI, Meta, Sakana, MiniMax, Z.ai, Moonshot AI, Thinking Machines, DeepSeek. A provider with no models available to you is left out entirely rather than shown empty.
Within a group, the newest release leads. The order is curated, not alphabetical or price-sorted, so the model you most likely want is the one nearest the group heading.
When a provider ships a successor, the previous release moves behind that provider’s Older versions row instead of leaving the list. Click the row to expand it. This keeps each group short enough to scan while keeping every model reachable — nothing is removed from the picker just because it was superseded.If the model you are currently using is an older version, its group opens already expanded, so your active selection is always visible.
Recently added models carry a green NEW badge. It is deliberately limited to the newest few releases rather than everything added since the last refresh — if most rows were badged, the badge would stop meaning anything. A model that your plan cannot select shows a lock instead of the badge.
Choosing an older version costs exactly the same as it did before it was superseded — the Older versions grouping is about keeping the list readable, not about deprecation or pricing. Removal from the API is announced separately in the changelog.
Fugu Ultra delivers output in larger batches rather than token-by-token — it thinks first, then sends a complete response. The chat header shows a notice when Fugu Ultra is selected.

Choosing a model

Choose GPT-6 Astra for OpenAI’s deepest reasoning and agentic coding — it has two independent controls, a thinking-effort level and a reasoning mode, so switch it to Pro when a hard problem is worth more model work and a slower answer. Or pick Claude Opus 5 for frontier Anthropic reasoning and coding. GPT-5.6 Sol remains available for frontier GPT reasoning at a lower cost tier. Fugu Ultra (Sakana) is also available on paid plans for hard, long-horizon multi-agent work — it uses deep reasoning with a 1M-token context window and delivers output in batches. Under Older versions, Claude Fable 5 remains available to eligible paid users for deep reasoning and long-horizon agent work, and Opus 4.8, 4.7 and 4.6 are there if you have existing prompts tuned to a specific version.
Claude Sonnet 5 is the default and recommended for most users. It handles coding, analysis, and creative tasks well at moderate cost. GPT-5.6 Terra is a balanced GPT option for eligible accounts. Claude Sonnet 4.6 is still available under Older versions if you have existing prompts tuned for it.
GLM 5.3 Flash, GLM 5.3, GPT-5.6 Luna, Muse Spark 1.3, and MiniMax M3 use very few credits per message. The GLM models are especially good at reasoning tasks; MiniMax M3 excels at very long-context work. These models are great for iterative sessions where you send many messages. GLM 5.2 costs the same as 5.3 and is still selectable under Older versions.
Claude Haiku 4.5 is the fastest model. Use it for quick questions, formatting, or lightweight code generation where speed matters more than depth.

Using models via the Developer API

Backend integrations can send an optional model field on Developer API run-creating endpoints. Use Supported Models for the accepted ids, or see the Developer API Models guide for request examples and lifecycle policy.

Prompt caching

Most models support prompt caching, which reduces cost and latency on follow-up messages in the same thread. When caching is active, repeated parts of the conversation (system prompt, earlier messages) are served from cache at a reduced rate. Caching happens automatically. You don’t need to configure anything.

Credit costs

Credits are deducted based on actual token usage. The cost tier determines how many credits each message uses: Exact costs depend on the length of your message, the conversation history, and how much the model outputs. You can see your remaining credits in the sidebar.