The range of AI tools is growing at lightning speed: ChatGPT, Gemini, Claude, Mistral, open-source, proprietary, text, image, reasoning, multimodal... For those who do not work with AI on a daily basis, it is difficult to keep an overview. Which large language model (LLM) fits you best? More importantly, how do you make that choice?
Let’s start with the basics: what do we mean by an LLM? A Large Language Model is an AI system trained on massive amounts of text—but also on images, audio, video, and even software source code. It can complete sentences, answer questions, write text, generate code, or perform complex analyses. But not all models are created equal. Some models excel at generating outputs (like text or images), while others truly reason, combine information, and analyze complex scenarios.
We see roughly three levels of ‘intelligence’:
1. Generative models
The first LLMs that came on the market that are ideal for texts, images or videos. Think campaigns, blogs or social posts applications.
2. Reasoning models
These go a step further. They can analyze, code, support complex decisions and even ask counter questions. Think use for data analysis, strategy, software development or business cases or doing extensive research.
3. AGI on the rise
These are models that are going to be as smart (or even smarter!) than humans. This is the direction it is heading: AI teammates who independently perform different roles and tasks. A kind of team of digital colleagues that you manage.
A model such as GPT-4.5 excels in language and communication. Whereas GPT-o3 (OpenAI's reasoning model) is more suitable for developers or data analysts, business strategists and researchers.
Many organizations immediately buy licenses for the most popular tools such as Copilot, ChatGPT, Gemini, without knowing exactly what they want to do with them. But: the model should fit the task, not the other way around. A quick content task requires something different than a complex data request.
And that is exactly where it often goes wrong now: technology is embraced without first looking at which processes will benefit from it. Our advice: don't start with tools, start with tasks. What tasks do your people perform? Which ones are repetitive, time-consuming, or structurally lagging?
Only when you know that can you choose which type of Large Language Model (LLM) best suits your team or department.
1. Identify which tasks remain repetitive, time-consuming or unused.
2. Examine which of those tasks can be taken over (in part) by an LLM.
3. Choose the right model for the task.
A few examples:
- Writing text or answering emails? Use GPT-4.5.
- In-depth data analysis, coding or sketching complex scenarios? Then choose reasoning models such as OpenAI's o3 or Gemini 1.5 Pro.
- Combinations of tasks or multimodal output (text + image + data)? Then look for models that automatically switch between outputs.
LLMs are broadly applicable, as long as you know what you are using them for. Some concrete examples:
- Marketer: Campaign ideas, website texts, visuals (Output model, e.g. ChatGPT, Gemini).
- Data analyst: Scenario analysis, dashboarding, queries (Reasoning model, e.g. Claude Opus, GPT03 Turbo with tools).
- Finance specialist: Cash flow analysis, business case validation (Reasoning model).
- Developer: Code writing, debugging, documentation (Reasoning model, e.g. Gemini 2.5 Pro, GPT-o3, Claude opus 4).
- Researcher/policy advisor: Hypothesis generation, literature review, scenario comparison (Deepresearch Model with Geminin l + search, e.g. Perplexity AI).
- Category manager: Automatically generate dashboards and reports (Reasoning model or multimodal model, e.g. Gemini 2.5 Pro).
Want to get started with Generative AI in your organization? Then start with a Smart Task Scan. With this we map out which processes and tasks are suitable for Generative AI, and which type of model fits them best. In this way, you avoid wrong investments in tools that do not help your team move forward.
Get in touch with us. We’ll interview your team or department—no strings attached—and provide tailored advice on which models best fit your work.