2024 has proven to be a landmark in the world of Artificial Intelligence, with a series of exciting innovations and milestones reshaping industries, the way humans and machines work together, and setting new standards in many areas. Now, at the start of 2025, it is worth taking a moment to take stock of what has happened in the past year.

Several companies and dynamic start-ups have presented new tools that have made the use of Artificial Intelligence widespread, particularly in the fields of health, finance, commerce and education. The emergence of OpenAI o1 and GPT-4o, innovations from Google and Meta, and the entry of Apple Intelligence into the field all indicate that the wave of AI has indeed reached all sectors.

It may be a cliché, but the incredibly fast evolution of the world of AI has created not only exciting opportunities but also challenges. One of these new challenges is often the very ability to keep up with the emergence of new models, services, technologies, companies and announcements. The events of 2024 were therefore not only about innovation, but also about social adaptation – a period of learning and change.

The competition for Large Language Models (LLMs) has been reinvigorated in 2024, and the market has become more colorful than ever. While the GPT-4-0314 version previously set the standard for advancement, models such as GPT-4o and other research projects or standalone products have since reached or even surpassed this performance. Google added two significant milestones to the year: after Gemini 1.5 Pro, they launched Gemini 2.0 at the end of the year. The latter is not only twice as fast as its predecessor, but also features advanced multimodal capabilities, bidirectional streaming and more complex data interpretation. Meta AI has also launched Llama 3.2, which has quickly become popular among developers thanks to its open-source nature. Not only did the number of model parameters increase significantly, but multimodal features – such as visual question-and-answer – also made the model stand out.

In addition to the big players, “smaller” companies have also shown impressive results. Anthropic Claude 3.5 Sonnet, for example, has deepened AI awareness and fine-tuned reasoning, showing that innovation is not the exclusive preserve of tech giants. This diversity has perhaps become more and more characteristic of the industry in the past year, as the first innovators have been followed by others in the development of Artificial Intelligence models. So, let’s look at each quarter in more detail.

In the first quarter of 2024 (Q1), innovations have already emerged that have further enhanced the AI revolution. Google unveiled AI-powered “Circle to Search” and “multi-search” features that have made information search more intuitive. At the same time, Groq debuted the LPU Inference Engine, which significantly accelerated the running of open-source models. Also, in Q1, OpenAI launched the GPT Store (or GPTs), where users can create customized ChatGPT versions for their own needs. Another new product from OpenAI, Sora, was introduced as a technology for creating videos from text, which was able to generate impressive, high-definition videos even during early demos, although its actual release was delayed until the end of the year.

During the period, Lag-Llama also attracted considerable attention: it was the first model to offer open-source time-series forecasts, opening up new possibilities for industry analysis.  Overall, these innovations have further solidified AI’s position at the forefront of the technological revolution.

As for Q2, Meta’s Llama 3 (with 8B and 70B parameter versions) further expanded the local run options, and the Microsoft Phi-3 family of models was also introduced here, focusing on smaller, more cost-effective LLMs (SLM). Google continued its catch-up strategy with the launch of Gemma 2, which took forward a branch of Gemini research. DeepMind has also made its mark with AlphaFold 3, which has brought significant advances in biomolecular research, enabling scientists to better understand molecular structures and their interactions, opening new avenues for drug development and other biotechnology applications. This diverse innovation also illustrates how the field of application of Artificial Intelligence has broadened.

As for local execution, it was observed in 2024 that smaller, less energy-intensive models began to appear in increasing numbers. Microsoft, Meta and several other market players have also continuously developed solutions that can be run on a single machine or mobile device. For example, Large 2, introduced by Mistral in Q3, is particularly beneficial in areas where longer texts, documents or interviews need to be handled due to its 128k context window, and is specifically optimized for single-node, i.e. local execution. Also worth highlighting from the third quarter is the announcement of OpenAI o1 and o1-mini. These are the first “reasoning” models that aim for higher-level logical inference and, in the case of the latter, faster response times.

Falling prices and changes in the hardware background remain key issues. NVIDIA continues to dominate the AI chip industry, thanks to its highly efficient GPUs and CUDA ecosystem, which it further strengthened in 2024 with the development of the Project GR00T base model. This move has further strengthened the field of robotics and embodied AI. At the same time, Stability AI has not been left idle: the Stable Video 3D solution announced in Q1 and later, in Q4, Stable Diffusion 3.5, both opened a new chapter in the generative field.

With better-performing models and a wide range of applications, artistic, creative and industrial tasks can increasingly be performed automatically or semi-automatically. This also foreshadows key trends for the upcoming year.

István ÜVEGES, PhD is a Computational Linguist researcher and developer at MONTANA Knowledge Management Ltd. and a researcher at the HUN-REN Centre for Social Sciences. His main interests include the social impacts of Artificial Intelligence (Machine Learning), the nature of Legal Language (legalese), the Plain Language Movement, and sentiment- and emotion analysis.