New — onFlow: your most critical process, automated with AILearn more
Perspective

There's no single AI: one model for each task

  • September 17, 2026
  • By Germán Middi
There's no single AI: one model for each task

Beyond chatbots, there are models specialised in music, images, data and engineering

For many people, "using AI" means opening ChatGPT and typing a question. It's a good starting point, but it's a small part of what's out there. Behind the general assistants, an ecosystem of specialised models has grown, each trained to do one thing well.

Music: Suno

Suno generates complete songs, with vocals and instruments, from a description. This month it released version 6, built with record labels and trained on licensed catalogues. For a business it's useful for a video's soundtrack, a jingle or hold music, without paying separate rights.

Images

  • GPT Image (OpenAI): the best at following long, detailed instructions.
  • Nano Banana (Google): very good at editing existing photos, with a free version.
  • Midjourney: the benchmark when style matters.
  • Ideogram: the best at writing text inside the image, useful for graphics and ads.

Data

Data platforms have added assistants that answer plain-language questions about the company's information: Genie in Databricks, Cortex in Snowflake, Copilot in Power BI. Instead of requesting a report and waiting, someone asks "which customer cut their purchases the most this quarter?" and gets the answer with the chart. For a single file, uploading it to ChatGPT or Claude is enough.

Engineering

In design and simulation, companies like Siemens, Ansys and NVIDIA have models that propose parts from performance targets or predict a simulation's result in seconds, without running the full calculation. They still need an engineer to validate, but they cut a lot of back and forth.

Development

At onMedia we use Claude Code and Codex every day to write code. They're general models, but connected to the project's code they perform in a way a standalone chat can't.

How to choose

The useful question is what concrete task you want to solve, and then which model is best for it. Often the answer isn't a general assistant, and sometimes it's a combination of several.

Is there a task in your company you'd like to solve with AI but don't know where to start? Get in touch and we'll look at it together.

Germán Middi
Written byGermán MiddiFounder at onMedia — Software engineer, automation & applied AI.

I build digital systems that help businesses run better. From Argentina to Australia, I work with companies turning complex operations into scalable platforms. 🚀 Laravel • Vue • AI workflows • Digital transformation Let’s connect if you’re building something meaningful.

You Might Also Like

Would you like to share your thoughts?

Your email address will not be published. Required fields are marked *