The History of AI Models - Part 4: Multimodality and Autonomous Agents (2023 - 2026)
In the final part of our journey into AI history, we examine today's multimodal models, the open-source revolution, and autonomous AI agents.
In the third part of our article series, we examined the invention of the Transformer architecture and how ChatGPT took the world by storm. ChatGPT created a massive impact, but fundamentally it was still a system that only understood and generated “text”.
In this final part of our AI history series, we discuss the process spanning from 2023 to the present year 2026, covering multimodality, the open-weight revolution, and the rise of autonomous agents.
2023: Multimodality and GPT-4
The human brain learns not just by reading, but by seeing, hearing, and touching. The next goal for AI researchers was to make models “Multimodal” in exactly this way.
In March 2023, OpenAI announced the GPT-4 model. Unlike its predecessors, GPT-4 could analyze not only texts but also images. For example, when you gave GPT-4 a sketch of a website drawn on a napkin, it could write the working HTML/CSS codes of that sketch in seconds.
Following this, Google took multimodality a step further with the Gemini series. Gemini models were designed from the ground up to process audio, video, image, and text data simultaneously (natively multimodal).
The Open-Weight Revolution and Llama
In 2023 and 2024, Meta (formerly Facebook) made a very critical move against the proprietary models of tech giants (OpenAI, Google): It opened the Llama model series for use by researchers and developers.
Thanks to this approach called “Open-Weight”, developers all over the world had the chance to run powerful AI models on their own computers or company servers. Especially the completely free release under the Apache 2.0 license of massive models with 295 billion parameters like Tencent Hunyuan 3 (Hy3) introduced in July 2026, dealt a massive blow to closed-system monopolies.
Now, companies can run LLMs locally on their own servers (on-premise) ensuring corporate data security without having to send their data to the cloud.
2025 - 2026: From Chatbots to Autonomous Agents
The biggest revolution of the current period (2025-2026) is Artificial Intelligence Agents (AI Agents).
Previous models were passive systems that “answered if you asked”. Autonomous agents, on the other hand, are systems that can make plans on their own to achieve a specific goal, do research on the internet, use various software tools (APIs, terminals, etc.), and learn from their mistakes.
At the point we have reached today, enterprise AI agents can work integrated with companies’ ERP systems, act as autonomous coding assistants to developers (like Cursor), and manage long-term complex projects for hours without human intervention.
Conclusion
The journey that began in the 1950s with Alan Turing’s question “Can machines think?” has been a long marathon stretching from simple perceptrons to AI winters; from the deep learning revolution to Transformer architectures. Today, we are not just facing computers that calculate, but intelligent digital colleagues that read, see, draw, write code, and can make decisions.
The evolution of artificial intelligence models continues unabated, and it is exciting even to imagine what the coming years will bring us.
We will continue to produce content on the development of artificial intelligence and its practical uses for businesses on the Canary Digital blog. Stay tuned!