Supported AI models
17+ models across OpenAI, Google, Anthropic, Meta, DeepSeek, xAI, and more.
Last updated: 2026-05-05
The model lineup
Pranas connects to 18 AI models across the major providers. We connect through OpenRouter (multi-vendor routing) plus direct integrations where they matter (for example, Google for AI Overview).
| Provider | Models |
|---|---|
| Google AI Overview (Gemini 3.5 Flash-Lite with Google Search), Gemini (consumer Gemini app, 3.8 Flash), Gemini 3.1 Pro (Preview) | |
| OpenAI | ChatGPT (consumer default, GPT-6 Luna), GPT-6 Sol, GPT-5.6 Sol, ChatGPT Pro (GPT-6 Astra) |
| Anthropic | Claude Haiku 4.5, Claude |
| Meta | Llama 4 Scout, Llama 4 Maverick |
| DeepSeek | DeepSeek V4 Flash, DeepSeek V4 Pro |
| xAI | Grok 4.7 |
| Mistral | Mistral Large |
| Perplexity | Sonar, Sonar Pro, Sonar Reasoning Pro |
The list changes as providers release new models and retire old ones. When a provider retires a model, we replace it with the current version. Your past scans keep their original model name, so old and new results are never mixed. The model picker in any scan dialog always shows what you can use today.
Tiers
Models are grouped into tiers based on cost and capability:
- Premium (ChatGPT Pro, GPT-5.6 Sol, Gemini 3.1 Pro): the strongest reasoning and recall, at the highest cost.
- Mid (GPT-6 Sol, Claude, Grok 4.7, Sonar Pro): a strong balance of quality and cost.
- Light (ChatGPT, Google AI Overview, Claude Haiku 4.5, Llama 4, DeepSeek V4 Flash, Sonar): fast and low-cost, good for breadth.
Use the mix that matches how your customers actually search.
Web-grounded vs training-only
| Mode | Models | What it means |
|---|---|---|
| 🌐 Web-grounded | ChatGPT, GPT-6 Sol, GPT-5.6 Sol, ChatGPT Pro, Gemini, Google AI Overview, Perplexity Sonar models | The model runs live web searches before it answers (Gemini and Google AI Overview use Google Search). Results reflect what consumers see today, including new brands, regional players and breaking news. Source URLs in your scan are real citations; Gemini scans don't return source links yet. |
| 🧠 Training-only | Gemini 3.1 Pro, Claude, Grok, Llama, DeepSeek, Mistral | The model answers from its training data, which is usually 6–12 months old. It is faster and cheaper, but may miss new or regional brands. |
Because these modes differ, different models can give different results for the same prompt. That is how each model behaves for real users. We show both faithfully so you see the full picture.
The 🌐 Web badge in the model picker shows which models are web-grounded.
Health monitoring
Every model is health-checked every 6 hours. If a provider retires a model or it keeps failing, the model is switched off automatically so you don't spend credits on it. If a scan can't reach a model at all, it is marked Failed rather than "No brands found", and no credits are charged for it.
Why we use OpenRouter
- One billing relationship, many providers
- Automatic failover and routing
- Faster access when new models are released
Google AI Overview searches Google for every answer and lists the exact pages it used in the Sources tab. Scans run before 15 October 2026 answered from the model's own knowledge; results continue under the same name when the model is upgraded, so charts show one combined line.