Model choice
An AI chatbot with GPT, Claude and Gemini: pick the model for each agent
Every agent answers with GPT-6 Luna from OpenAI by default. On a paid plan you can switch an agent to a model from Anthropic or Google, or let Auto pick one for each message, and you see exactly how many messages each reply uses.
One model rarely fits every conversation
A greeting, a shipping question and a three-part complaint about a refund are not the same task. A fast model handles the first two well and cheaply. A larger model writes a more careful answer to the third, but costs more per reply. Many chatbot builders decide this for you and hide the cost. intoCHAT lets you choose per agent, shows the price of each choice in messages before you switch, and keeps everything else about your agent the same.
Why choose the model yourself
Six models from three providers, one setting per agent, and the rest of your agent unchanged.
Three providers in one dropdown
GPT-6 Luna from OpenAI, Claude Haiku 4.5, Claude Sonnet 5.5 and Claude Opus 5.5 from Anthropic, and Gemini Flash and Gemini Pro from Google. intoCHAT connects to each provider, so you have no API keys of your own to manage.
The cost is on screen before you switch
Each model shows its tier and how many of your plan's messages one reply uses: 1 for the fast and standard models, 2 for Claude Sonnet 5.5 and Gemini Pro, 5 for Claude Opus 5.5.
Auto picks a model for every message
Auto sends simple messages to the fastest model and complex ones to GPT-6 Luna. Every reply with Auto uses one message, whichever model writes it.
Visitors always get a reply
If a provider fails before the reply starts, GPT-6 Luna answers instead, and the extra messages of the larger model are given back, so the reply uses one.
Nothing else about your agent changes
Knowledge, instructions, guardrails, actions, forms and procedures work the same with every model. Different agents in your account can use different models.
How to choose a model
Open the Model card
Open your agent, go to the Setup tab and find the Model card below Reply language. Each model lists its provider, its tier, its messages per reply and the cheapest plan that includes it. Models your plan does not include are greyed out with the plan they need, such as "Needs Pro".
Try it in the Playground first
Pick a model and chat in the Playground before you save; it uses the model on screen. In Compare, side A and side B can each use their own model, so you can read two answers to the same question side by side.
Save, then check the results
Click Save changes. The Models used card on the Dashboard shows which models wrote your replies, Why this answer under each reply in a transcript names the model and any fallback, and Billing lists replies by model for the current period.
The models and what one reply costs
GPT-6 Luna (OpenAI), the default
Standard tier, 1 message per reply, on every plan including Free. Fast and capable for most support conversations. Reads images and PDFs that visitors attach.
Claude Haiku 4.5 (Anthropic)
Fast tier, 1 message per reply, from Starter. The quickest replies, for simple questions at high volume.
Gemini Flash (Google)
Fast tier, 1 message per reply, from Starter. Fast and inexpensive, and good with long knowledge.
Claude Sonnet 5.5 (Anthropic)
Advanced tier, 2 messages per reply, from Pro. Careful, well-written answers and reliable tool use.
Gemini Pro (Google)
Advanced tier, 2 messages per reply, from Pro. Strong reasoning over large amounts of knowledge.
Claude Opus 5.5 (Anthropic)
Premium tier, 5 messages per reply, from Pro. The most capable, for complex questions and multi-step actions. Ten replies use 50 messages.
Auto
1 message per reply, from Starter. Chooses between Claude Haiku 4.5, Gemini Flash and GPT-6 Luna for each message, and only from the models your plan includes.
Background AI stays on OpenAI
Searching your knowledge, topic analysis, spam checks, the copilot, Ask your conversations, live translation and the conflicting facts check always use intoCHAT's standard OpenAI model, whatever you pick.
Frequently asked questions
Which AI models can my chatbot use?
GPT-6 Luna from OpenAI is the default and is included on every plan, including Free. From Starter you can also pick Claude Haiku 4.5, Gemini Flash or Auto. From Pro you can pick Claude Sonnet 5.5, Gemini Pro and Claude Opus 5.5 as well. Enterprise includes every model. A model only appears in the list while intoCHAT has its provider set up.
How many messages does a reply use with each model?
GPT-6 Luna, Claude Haiku 4.5 and Gemini Flash use 1 message per reply. Claude Sonnet 5.5 and Gemini Pro use 2. Claude Opus 5.5 uses 5, so 10 replies use 50 messages. Auto always uses 1. A reply that runs actions or a web search still counts once, at its model's rate. On Pro, with 10,000 messages a month, that is 5,000 replies with Sonnet or 2,000 with Opus. An agent's monthly message cap counts the same way.
How does Auto decide which model answers?
No model is asked to decide. Simple rules read each message instantly, so Auto adds no delay and uses no extra messages. Greetings, thanks, yes or no answers and short questions go to the fastest model available: Claude Haiku 4.5, then Gemini Flash, then GPT-6 Luna. GPT-6 Luna takes messages with an image or file, questions in several parts, messages over 400 characters, comparisons, requests for code or a table, requests such as a refund or a booking when the agent can act on them, and any message once a conversation has more than 12 earlier messages. The rules understand English, German, French, Italian and Spanish.
What happens if Anthropic or Google has an outage?
If the provider fails before the reply starts, intoCHAT tries GPT-6 Luna once instead and the rest of that reply stays on it. The reply then uses one message: the extra messages of the larger model are given back. intoCHAT waits up to 25 seconds for Anthropic or Google before it falls back. A reply that is already streaming cannot switch models, so in rare cases a visitor may see a reply stop short.
Are there other cases where the default model answers?
Yes, and in each of them the reply uses one message. GPT-6 Luna answers when your plan no longer includes the chosen model, for example after a downgrade from Pro to Starter; when the model is no longer set up on intoCHAT; and when fewer messages are left this month than the model uses, but at least one. Your settings are never changed: the Model card tells you that replies use GPT-6 Luna until you upgrade or pick another model.
Does the model change what my chatbot knows or can do?
No. Knowledge, instructions, guardrails, actions, forms and procedures stay the same. The model writes every reply the agent gives: in the chat widget, the inline iframe and chat link, the Help Page, Slack, email, the REST API and the MCP server. Answers do differ in style and length, and larger models usually take longer to answer, so try a model in the Playground before you switch a busy agent. Every model reads images visitors attach; PDFs are read by OpenAI models only.
Where do visitor messages go when I pick Claude or Gemini?
The visitor's messages, the conversation so far, your instructions and the knowledge excerpts used for the reply go to the provider of the model that writes it: OpenAI, Anthropic or Google. intoCHAT's background work, such as searching your knowledge and spam checks, always uses OpenAI, so messages also reach OpenAI whichever model you pick. If your privacy policy names the companies that process chat messages, list OpenAI and the providers of the models you use.
Pick the right model for each agent
Start free on GPT-6 Luna. Starter adds Claude Haiku 4.5, Gemini Flash and Auto; Pro adds Claude Sonnet 5.5, Gemini Pro and Claude Opus 5.5.
Create your agent freeFree plan, no credit card needed.