Skip to main content

​
What are AI Models?

AI models are the engine behind your AI Agent. They process input (like customer questions) and generate output (like answers). Each model has unique strengths in reasoning, speed, memory, and language capabilities. Understanding these differences helps you build a more effective AI Agent.

​
How to choose the best model for your Agent?

1

Define your AI Agent’s primary function

Is it answering FAQs, guiding users through tasks, or automating actions?
2

Set your priorities

What is the most important thing the Agent should do?
  • Speed: Do you want fast replies, or can the Agent take some more time?
  • Length of answers: Does the Agent need to give long answers, or quick and short?
  • Sources: How many sources does the Agent need for answering questions?
3

Pick the best model for your use case

Use the table below to find the best model for your situation:
Use caseExamplesRecommended modelGuidance
Basic conversationsFAQ questions, delivery status checks, booking confirmationsGPT 5.4 NanoUse for simple, high-volume conversations where speed matters most. This is the fastest AI Engine 2.0 model and is ideal for repetitive questions that can be answered from clear knowledge base content.
Longer or more complex customer journeysInsurance claims, onboarding processes, resolving complex questions across multiple messagesGPT 5.4 MiniUse for customer journeys that require more context, better interpretation, and higher answer accuracy across multiple messages.
Smart, complex conversationsTroubleshooting, legal compliance, internal IT policies, complex customer supportGPT 5.4 MiniUse when reliability and reasoning are more important than maximum speed. Recommended for conversations where the Agent needs to combine multiple sources or handle nuanced questions.
Internal Support AgentsHR, IT, and finance questions, leave requests, password resets, payroll explanationsGPT 5.4 MiniUse for internal support where policy accuracy and clear explanations matter. For very simple, high-volume internal FAQs, GPT 5.4 Nano can also be used.
Multilingual supportAgents that switch smoothly between Dutch, English, German, French, Spanish, or other languagesGPT 5.4 MiniUse for multilingual organizations where answer quality and reliable retrieval across languages are important. AI Engine 2.0 is designed to find the right knowledge more consistently, even when the question and knowledge base are in different languages.
Agent needs to know date and timeReservations, new launches, pricing changes, policy updates, availability questionsGPT 5.4 MiniUse when the Agent needs to interpret time-sensitive information carefully. Make sure the relevant website, knowledge base, documentation, or product feed has been re-crawled so the Agent uses the latest available knowledge.
Agent needs to execute ActionsTriggering workflows, creating CRM entries, booking meetings, checking order status, performing multistep automationsGPT 5.4 MiniUse for Agents that work with Actions or integrations. AI Engine 2.0 supports more dynamic tool choice, allowing the Agent to select the most relevant action based on the customer’s request. Add clear instructions that define when the Agent is allowed to trigger actions and when it should ask for confirmation.
Large product catalogsWebshops with many similar products, product comparisons, product availability questionsGPT 5.4 MiniUse when the Agent needs to distinguish between similar products, product names, variants, or specifications. AI Engine 2.0 improves entity recognition, helping the Agent understand exactly which product the customer means.
Many SKUs or technical identifiersSKU comparisons, order numbers, product codes, locations, company names, technical referencesGPT 5.4 MiniUse when precision is important. AI Engine 2.0 recognizes specific entities such as SKUs, product names, order numbers, company names, locations, and function names, which reduces confusion and improves answer accuracy.
High-volume automationFirst-line customer support, ecommerce support, repetitive support questionsGPT 5.4 NanoUse when the goal is to automate a large number of straightforward conversations as quickly as possible. Nano is the best fit when speed and scalability are the main priorities.
Accuracy-critical supportCompliance-sensitive questions, technical documentation, API documentation, policy explanationsGPT 5.4 MiniUse when incorrect answers create higher risk. Mini benefits most from AI Engine 2.0 improvements such as Hybrid Search, reranking, improved RAG, better source selection, and smarter follow-up questions.
Model recommendation note
AI Engine 2.0 introduces GPT 5.4 Nano and GPT 5.4 Mini as the recommended models. Other models may still be available in the platform, but for new and upgraded Agents, customers should be guided toward these two models because they are optimized for AI Engine 2.0.Use GPT 5.4 Nano when speed, scale, and simple automation are the main priorities.Use GPT 5.4 Mini for most other use cases, especially when conversations are more complex, the knowledge base is larger, accuracy matters more, or the Agent needs to use Actions.
If you choose GPT-5 as your model, it’s essential to include an instruction in your Agent’s setup that clearly states it is not allowed to independently execute tasks or take actions.

GPT-5 has advanced reasoning and autonomy, so without this kind of instruction, it may try to take initiative where it shouldn’t.Example instruction: “You are not allowed to take actions, complete tasks, or trigger workflows.”

​
Missing your use case in the above schedule?

Compare the different models below to find the best match. The questions below can help you.
  • Do you need fast responses? Go for a mini or nano version of your preferred model
  • Do you want the Agent to be able to handle a lot of information? Use a model with a high context window.
  • Do you want the Agent to provide detailed answers? Go for a model with high output tokens.
  • Do you want the Agent to think well, make connections, and solve complex tasks? Go for an Agent with high reasoning.
ModelReasoningSpeedInputOutputContext WindowMax Output TokensKnowledge Cutoff
GPT-5.4-miniHigherFastText, ImageText400,000128,000Aug 31, 2025
GPT-5.4-nanoHighFastText, ImageText400,000128,000Aug 31, 2025
GPT-5.1HighestMediumText, ImageText500,000200,000Dec 01, 2024
GPT-5HigherMediumText, ImageText400,000128,000Oct 01, 2024
GPT-5-miniHighFastText, ImageText400,000128,000May 31, 2024
GPT-4.1HigherMediumText, ImageText1,047,576128,000Jun 01, 2024
GPT-4.1-miniHighFastText, ImageText1,047,57632,768Jun 01, 2024
GPT-4oHighMediumText, ImageText128,00016,384Oct 01, 2023
o3HighestSlowestText, ImageText200,000100,000Jun 01, 2024
o3-miniHigherMediumTextText200,000100,000Oct 01, 2023
You can always test different models and compare them using the Playground.