# Which AI Provider Can Do What?

> **What it does:** One page to answer "why is this feature greyed out / erroring?" — every feature × every provider, plus how to combine providers smartly.
>
> *Applies to Antradus AI Lite & Pro v2.0.0 · Last updated 2026-07-24*

## The matrix

| Capability | OpenAI | Anthropic (Claude) | Google Gemini | OpenRouter | DeepSeek |
|---|---|---|---|---|---|
| **Article writing** | ✅ | ✅ | ✅ | ✅ 200+ models | ✅ |
| **Image generation** (featured + albums) | ✅ gpt-image / DALL·E | ❌ no image API | ✅ Imagen | ✅ Flux & compatible | ❌ |
| **Vision** (product-from-photo) | ✅ | ✅ | ✅ | ⚠️ model-dependent | ❌ |
| **Live web search** | ⚠️ search-capable models only (gpt-5 / gpt-4o) | ✅ | ✅ Google grounding | ✅ any model | ❌ |
| **Free tier models** | ❌ | ❌ | ❌ | ✅ ★ FREE labels | ❌ |
| **In-editor assistant / transforms** | ✅ | ✅ | ✅ | ✅ | ✅ |
| **Bulk publishing** | ✅ | ✅ | ✅ | ✅ | ✅ (no web search, no images) |

## Practical pairings

- **"I want everything from one key"** → OpenAI or Gemini (writing + images + vision + search).
- **"Best writing, images secondary"** → Claude for text; note you'll have no image generation — pair with a second provider when you need images.
- **"Minimum cost"** → OpenRouter with a ★ FREE model for writing; expect variable quality, and remember free-model providers often retain/train on prompts.
- **"Maximum privacy of drafts"** → any paid API tier (paid API traffic is generally excluded from training by the major providers — check your provider's current terms).

## Switching providers

Keys are stored per provider — switching the **Active Provider** in settings is instant and non-destructive. A common pro setup: bulk-write a batch on a cheap/free model, then regenerate the most important articles on a flagship model.

## Where the model choice shows up

Every generation uses whatever Model / Image Model is selected *at that moment* — articles, meta, transforms, bulk jobs, images. The [Reporting page](../bulk/reporting.md) records usage per model so you can compare real costs.
