Social Intents supports the three major AI providers - OpenAI ChatGPT, Anthropic Claude, and Google Gemini (directly or through Google Vertex AI) - and you can switch between them at any time from the AI Agent tab without retraining anything. Here is how they differ for customer-facing chat, which models are included with your subscription, and how to pick. Each engine also has an overview page: ChatGPT chatbot, Claude chatbot, and Gemini chatbot.
The Short Answer
All three produce excellent customer-facing answers when trained on good content. If you have no preference, keep the default, GPT-6 Luna: it is included with your subscription (no API key), fast, and has a 1M token context window. Switch engines later if you have a reason to; your knowledge, instructions, and actions carry over unchanged.
Current Models
Pick the model under Chat Widgets > your widget > AI Agent > Agent settings > Provider & credentials.
| Provider | Models in the dropdown | API key |
|---|---|---|
| ChatGPT (OpenAI) | GPT-6 Luna (default, 1M context), GPT-5.6 (best quality), GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.2, GPT-4.1, GPT-4.1 Mini, GPT-4o, GPT-4o Mini, GPT-4 | Included: GPT-6 Luna, GPT-5.6 Luna, GPT-4o and GPT-4o Mini. All other ChatGPT models use your own OpenAI key. |
| Anthropic Claude | Claude Sonnet 5, Claude Opus 4.6, Claude Sonnet 4.6, Claude Haiku 4.5, Claude Sonnet 4.5 | Always your own Anthropic API key. |
| Google Gemini | Gemini 3 Pro (preview), Gemini 3 Flash (preview), Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash Lite, Gemini 2.0 Flash | Always your own Google API key. |
| Google Vertex AI | Gemini 2.5 Pro, 2.5 Flash, 2.5 Flash Lite and 2.0 Flash through your Google Cloud project | Uses your Vertex region and project ID instead of an API key. |
How They Compare for Live Chat
| Consideration | ChatGPT (OpenAI) | Claude (Anthropic) | Gemini (Google) |
|---|---|---|---|
| Response speed | Fast, especially the Luna and Mini models | Fast on Sonnet and Haiku; Opus trades speed for depth | Fast, especially Flash models |
| Answer style | Direct and versatile | Careful, well-structured, strong at following instructions precisely | Concise, strong factual grounding |
| Long/technical content | Strong; most current models have a 1M token context | Particularly strong with long documents and nuanced policies | Strong |
| API key | Four models included; others use your key | Your own Anthropic key (no included models) | Your own Google key or Vertex project (no included models) |
Picking a Model Tier
Within each engine you choose a model tier. The tradeoff is the same everywhere:
- Fast/light models (GPT-6 Luna, GPT-5.6 Luna, GPT-4o Mini, Claude Haiku 4.5, Gemini Flash models) - quick responses and lower cost, great for FAQ-style support. This is the right default for most chat widgets.
- Flagship models (GPT-5.6, Claude Opus 4.6 or Sonnet 5, Gemini 3 Pro) - better reasoning on complex, multi-step questions and long policies, at somewhat slower speeds. Use them when your questions are genuinely hard: technical products, compliance-heavy answers, multi-condition pricing.
Visitors weigh speed heavily - a great answer in one second beats a slightly better answer in eight. Start light, and upgrade the tier only if you see reasoning failures in transcripts.
Using Your Own API Key
Paste your provider key in the API key field under Provider & credentials to use any model that is not included, such as GPT-5.6 or any Claude or Gemini model. Usage then bills directly to your provider account. Teams do this to use a specific flagship model, to consolidate AI spend, or to satisfy procurement requirements. Your key is stored securely and never shown again after saving; leave the field blank to keep it, or click Remove key to delete it.
Switching Engines Safely
- Change provider and model in the AI Agent tab (or in the Playground's Model & response style panel) and save.
- Open the Playground and rerun your standard test questions - tone and phrasing will shift slightly between engines.
- Adjust the instructions if needed (each engine responds a little differently to the same prompt) and save.
Training data, Q&A pairs, AI actions, and handoff settings all work identically across engines - nothing else needs to change.
Frequently Asked Questions
Which engine gives the most accurate answers?
Accuracy is dominated by training quality, not engine choice. A well-trained knowledge base on any engine outperforms a poorly trained one on the "best" model. Invest in training content first.
Which models can I use without an API key?
GPT-6 Luna, GPT-5.6 Luna, GPT-4o and GPT-4o Mini. They run on the key included with your subscription. Claude and Gemini always run on your own key.
Can different widgets use different engines?
Yes - the engine is a per-widget setting. Run Claude on your technical docs site and ChatGPT on your marketing site if you like.
Do I pay for AI usage?
Included models draw from your plan's AI response allowance. With your own API key, token usage bills to your provider account instead. See Plans, Billing & Upgrades.
What happens if my chosen model is retired?
When a provider retires a model, widgets that used it are moved to a comparable supported model, and the dropdown always reflects the current list. Check the AI Agent tab after a provider announcement and retest in the Playground.