Skip to main content
Omnifact supports multiple model families from top providers like OpenAI, Anthropic, Google, and Mistral. The “best” model isn’t always the most powerful one—it depends on your specific task, speed requirements, and hosting preferences. You can switch models directly in the chat interface at any time to find the one that works best for your current needs.

​
How to switch models

You can select the AI model before you send a message or at any point during a conversation using the model selector in the chat header. You can also switch models using the model selector in the chat input area. See Selecting an AI Model for more details. For details on which models are available to your team, see Model Management.

​
Model Tiers & Access

Models in Omnifact are categorized into three access tiers. Your ability to use certain models depends on the permissions assigned to your user group by your team administrator. For full details on managing these permissions, see Model Management. Choose your model based on your task’s requirements and your access tier:

​
Premium Tier

Best for: Complex reasoning, advanced coding architecture, nuanced analysis, and multi-step problem solving. These models are significantly more expensive than standard tier models, and access is managed via group permissions.

GPT-5.6 Sol

OpenAI’s most capable model for highly complex reasoning.

Claude Opus 4.8

Anthropic’s highest capability tier for complex reasoning.

o3

Reasoning-focused models that “think” before responding.

Gemini 3.5 Flash

Google’s top-tier model with strong reasoning capabilities across modalities.

​
Standard Tier

Best for: Complex writing, everyday coding, balanced performance, and high-quality output. Access to these models is managed via group permissions.

GPT-5.6 Terra

OpenAI’s highly capable model offering a balance of performance and speed.

Claude Sonnet 5

Anthropic’s balanced model, excellent for writing and coding tasks.

o3-mini

Next-generation efficient reasoning models.

Mistral Large

Mistral’s highly capable model, offering strong performance and EU hosting options.

​
Base Tier

Best for: Quick questions, summaries, simple tasks, and speed. Available to all users by default.

GPT-5.6 Luna

Fast and cost-effective for everyday tasks.

Claude Haiku 4.5

Extremely fast and efficient, great for simple queries.

Gemini 3.1 Flash Lite

Optimized for high speed and low latency.

Mistral Small

A capable small model for general use.

​
Image Generation

Best for: Creating visual assets from text descriptions. Your team administrator enables image models and sets the default in Team Settings > Models. The chat model selector does not switch image generation models.

​
OpenAI Models

GPT-Image-2

OpenAI’s latest model for fast, high-quality generation and editing. The default when OpenAI is configured.

GPT-Image-1.5

Strong prompt adherence and image quality for polished creative work.

GPT-Image-1

High-fidelity general-purpose image generation.

GPT-Image-1 Mini

Cost-effective option balancing speed and quality for everyday visuals.

​
Google Models (Nano Banana)

Nano Banana 2

Fast and efficient model for quick iterations and responsive creative workflows.

Nano Banana Pro

Advanced model for richer compositions and more demanding creative tasks.

Nano Banana

Balanced speed and quality for general-purpose image generation.
Some older models may still appear in your workspace for compatibility. For new work, prefer the newer GPT-5, Claude 4, Gemini 2.5/3, and Mistral models.

​
Strategy: Combining Models

You don’t have to stick to one model for an entire conversation. A powerful workflow involves combining models:
  1. Draft with Speed: Use a Base tier model like GPT-5 mini or Claude 4.5 Haiku to create outlines, brainstorm ideas, or draft initial text. This is quick and saves quota.
  2. Refine with Quality: Switch to a Standard tier model like GPT-5.2 or Claude 4.6 Sonnet to rewrite, polish, or critique the work.
  3. Solve Hard Problems with Reasoning: If you hit a logic blocker, switch to a Premium tier model like o3 or Claude 4.6 Opus to work through the specific problem, then switch back.
Common Pitfall: Many users stick to the default model for everything. While capable, you might be using “sledgehammer” models for simple tasks (wasting quota/speed) or struggling with complex tasks that a reasoning model could solve easily.

​
How to read model names

Model names often follow patterns that hint at their capability and speed.
While “4o” in GPT-4o typically indicated a premium omnichannel model, naming conventions can vary by provider. Always check the model description if unsure.

​
Model Hosting, Privacy, and Compliance

The Privacy Filter protects you regardless of the model.Omnifact’s Privacy Filter adds a layer of protection to every interaction, masking sensitive data before it reaches any model, regardless of where that model is hosted.
When choosing a model, consider your data residency requirements:
  • EU-hosted by default: Azure OpenAI, Mistral AI
  • US-hosted: OpenAI, Anthropic, Google
  • Custom providers: Hosting depends on your specific workspace configuration (e.g., Google Vertex AI, Anthropic via Vertex AI, AWS Bedrock, or other Custom providers).
In the model picker, look for the hosting-region indicator flag to confirm where a model is processed.

​
Image Generation Restrictions

When the Privacy Filter is enabled, image generation may be restricted to specific models or providers that comply with your organization’s data privacy standards (often requiring EU hosting). If you cannot select certain image models, this compliance setting is likely active.