👾AI models

The article discusses artificial intelligence models on the basis of which you can create your own virtual managers.

The Selarti service offers ten AI models: GPT-4.1, GPT-4.1-mini, GPT-4.1-nano, GPT-4.5 preview, o1, o3-mini, GPT-4o, GPT-4o-mini, GPT-4 Turbo and GPT-4 Turbo Preview. These models differ in functionality, performance and price, allowing you to choose the most suitable one for specific tasks.

This resource is an open source; data uploaded to it become available to OpenAI. There is no encryption.

You can read the differences between the models on the official OpenAI website, or you can read below in the article. You can also review Selarti's pricing here.

Comparative table of AI models

Context window — the amount of text the model can view and process to generate a response. The larger it is, the more context from the dialogue the model will remember during a long conversation.

Name	Input price	Output price	Context window	Speed	Purpose	Ability to reason
GPT-4.1	$2.40	$9.6	1047576	Average	For complex dialogues, marketing, customer support, educational and business products	Above average
GPT-4.1-mini	$0.48	$1.92	1047576	Above average	For everyday tasks: chatbots, recommendations, cheap text generation	Average
GPT-4.1-nano	$0.12	$0.48	1047576	Very high	For simple tasks: autocomplete, filtering, classification, microservices	Below average
GPT-4.5 preview	$90.00	$180.00	No information yet	No information yet	For tasks where high accuracy is important: medicine, law, finance, technical texts	No information yet
o1	$18.00	$72.00	200000	Slow	For analytics, research, decision making, deep reasoning	Above average
o3-mini	$1.32	$5.28	200000	Average	For logical tasks, programming, data work, complex assistants	Above average
GPT-4o	$3	$12.00	128 000	Average	For multimodal tasks: voice, image, text. Suitable for universal assistants and video	Average
GPT-4o-mini	$0.18	$0.72	128000	Above average	For fast multimodal solutions: visual and voice AI bots	Below average
GPT-4 Turbo	$12.00	$36.00	128000	Average	For long dialogues, analytics, generation of documentation and instructions	Below average
GPT-4 Turbo Preview	$12.00	$36.00	No information yet	No information yet	For testing new capabilities, early access to improved features	No information yet

Input and output prices are indicated per 1 million tokens.

The table does not contain all the information about models available for preview access — the capabilities of some models are still being collected.

In the sections below we describe the features of the models in a bit more detail, and at the end of the article we share our experience and observations about creating the Ilyusha bot on the o3 and GPT-4o models.

GPT-4.1

GPT-4.1 is OpenAI's flagship artificial intelligence model designed for complex tasks that require deep contextual understanding, advanced coding skills and strict adherence to instructions.

GPT-4.1-mini

GPT-4.1-mini is a cheaper and faster variation of the previous model. It responds faster to requests but reasons slightly less. Well suited for chatbots.

GPT-4.1-nano

GPT-4.1-nano is an even cheaper and faster variation of GPT-4.1. It works quickly but reasons less well. The cheapest in the lineup.

GPT-4.5 preview

GPT-4.5 preview is a research preview of the most powerful GPT model for conversational applications. The model represents a new step in the evolution of AI, demonstrating improvements through scaling both unsupervised training and task-specific training methods.

Main differences from previous models:

• Deeper understanding of the world — scaling compute and data reduces hallucinations and increases the reliability of responses.

• Improved emotional intelligence (EQ) — new training methods enable better understanding of user intent and more natural dialogue.

• Extended capabilities — support for search, file and image uploads, as well as integration with APIs for complex tasks.

• Raised safety standards — implementation of new oversight and testing techniques improves quality control and model safety.

o1

OpenAI o1 is a new artificial intelligence model introduced by OpenAI in September 2024. The model's feature is that it can "reason like a human" and solve more complex programming and math problems.

Key characteristics:

• High accuracy and quality of text generation.

• Good at solving mathematical problems and their analysis.

o3-mini

OpenAI o3-mini is a new economical model for complex tasks in STEM, programming and mathematics.

Key characteristics:

• Improved reasoning — the model "thinks" before answering, which increases accuracy.

• Adjustable reasoning levels — choosing between low, medium and high modes allows balancing speed and quality.

GPT-4o

GPT-4o (GPT-4 Omni) is an optimized version of GPT-4. This model is designed to provide a better price/performance ratio while maintaining high performance and accuracy.

Key characteristics:

• Improved performance at lower cost.

• Optimization for specific tasks.

• Reduction of compute costs.

GPT-4o-mini

GPT-4o-mini — is an economical version of the GPT-4 model intended for performing simple tasks. Suitable for tasks that do not require high intelligence and large computational resources.

GPT-4 Turbo

GPT-4 Turbo is an accelerated version of GPT-4 that provides higher request processing speed without a significant compromise in quality.

Key characteristics:

• High speed of response generation.

• Support for a large volume of real-time requests.

• Maintaining high text quality.

GPT-4 Turbo Preview

GPT-4 Turbo Preview is a preliminary version of GPT-4 Turbo, providing access to the latest improvements and optimizations. This model allows users to evaluate new capabilities and performance before the official release.

Key characteristics:

• Access to the newest features and improvements.

• High speed and performance.

• Possibility of feedback for further optimization.

Our experience in creating a technical support bot on the o3-mini and GPT-4o models

For several months now we have been creating a bot for our onlinePBX project. This bot will answer customers' simple questions and transfer them to staff if the question is complex. By the way, we named it Ilya; since it's a technical support staff member, it should have a name.

In the process of creating the bot we used two models — o3-mini and GPT-4o. Below we share our observations.

Working with documentation:

• o3-mini refers to documentation even if forbidden to do so. It usually does this in a citation format that displays incorrectly. Sometimes it replaces quotation marks with emojis.

• o3-mini searches for topic mentions across all files and does this more attentively than GPT-4o.

• GPT-4o also references documentation; the prohibition does not always work.

• GPT-4o mentions uploaded files if they contain no information, for example it may say: "There is no such information in the uploaded files."

• Formatting a knowledge base in both cases requires a good structure — it is important to highlight items and subitems with headings so models can save time and resources by "reading" only what relates to the customer's specific question.

Model reasoning:

• If there is no clear answer to a customer's question in the documentation, GPT-4o may give either a very vague answer or say that such information does not exist. In the same situation o3 will analyze the information and make up an answer. Sometimes it will be correct, and sometimes completely fabricated.

• o3-mini has a larger context window, so it will process more information than GPT-4o.

• GPT-4o provides an answer quickly, usually within 10 seconds.

• o3-mini responds longer, depending on the set reasoning parameter — about 15 seconds at the minimum, 20–30 seconds at medium and about a minute at the high level.

Transferring the question to an operator:

• Both o3-mini and GPT-4o require clear instructions for handing over the dialogue to an operator, but for o3-mini this needs to be described much more thoroughly, because the model will assume it can resolve the issue itself and will invent answers that do not correspond to reality.

• Unlike o3-mini, GPT-4o can hand the dialogue to an operator if it has no information to answer. o3-mini does not understand the concept of lack of information and continues to fabricate.

Text formatting:

• GPT-4o uses Markdown markup, which includes bold, italic and link formatting not as raw URLs but as hyperlinks for better text perception. If the bot will be used through a helpdesk, all markup should be forbidden, otherwise all text styling and links will be displayed incorrectly.

• o3-mini outputs text in a simple structure, may break it into items, but will send links in URL format.

Other features:

• When composing a prompt (instruction) GPT-4o easily works with roles and acts in character, may use more "friendly" phrasing and convincingly play the role of a support agent.

• o3-mini is poor at handling roles; if asked to be more emotional and friendly it is more likely to use expressive punctuation, for example adding more exclamation marks. o3-mini is more like a dry reference manual than an empathetic agent.