Milestones in Modern LLMs

What are some major milestones in modern LLMs?

Large language models (LLMs) have evolved rapidly over the past several years. The Transformer architecture, introduced in 2017, laid the groundwork for the models that followed. BERT applied it to bidirectional language understanding in 2018, one of several early transformer-based models alongside GPT-1, while GPT-2 and GPT-3 demonstrated increasingly fluent text generation from 2019 to 2020. The release of ChatGPT in late 2022 introduced conversational interfaces to a mainstream audience and kickstarted widespread adoption.

Open-weight projects such as LLaMA and the rise of specialized instruction tuning accelerated progress in 2023 and 2024, alongside GPT-4 and GPT-4o, which added multimodal input and longer context windows.

2024 and 2025 brought a shift toward reasoning-focused models trained to work through problems step by step before answering, starting with OpenAI's o1 and followed by open releases like DeepSeek-R1. Anthropic's Claude 4 family and Google's Gemini 2.5 models extended this reasoning approach alongside stronger coding and agentic tool-use capabilities, and OpenAI's GPT-5 consolidated its reasoning and general-purpose models into a single system.

Milestones increasingly revolve around practical considerations—scalable deployment, agentic workflows, alignment with human intent, and responsible use—rather than pure model size. As LLMs integrate into products and workflows, their milestones reflect both technical breakthroughs and their impact on everyday applications. This timeline covers developments through 2025.