Calling 2025 “the year of AI model architectures” feels a bit like saying “you should add ‘Reddit’ to your Google search to get better results.”
It’s not wrong. It’s just… a little late to the conversation.
Here’s how long these model types have actually been around:
• LLMs – 2018 (GPT-2, BERT)
• MLMs – 2018 (BERT, the original bidirectional model)
• MoE – 2017–2021 (Switch Transformer, GShard)
• VLMs – 2020–2021 (CLIP, DALL·E)
• SLMs – 2022–2023 (DistilBERT, TinyGPT, Phi-2)
• SAMs – 2023 (Meta’s Segment Anything)
• LAMs – 2024–2025 (Tool-using agents, Gemini, GPT-4o)
• LCMs – 2024–2025 (Meta’s SONAR embedding space)
These aren’t new ideas. They’re rebrands of ideas that finally hit product-market-fit.