FLAN-UL2 vs. Llama

Overview

Similar to FLAN-T5, FLAN-UL2 is a model based on Google's popular T5 architecture with an upgraded pre-training procedure dubbed UL2. On most NLU benchmarks, FLAN-UL2 outperforms FLAN-T5 by a significant margin.

Initial release: 2023-03-03

Reference

https://huggingface.co/google/flan-ul2

Further Reading

• https://www.yitay.net/blog/flan-ul2-20b

Overview

Llama is Meta's open-weight model family. Llama 4 was its last major generation; Meta's active assistant-model development has shifted to Muse.

Meta introduced LLaMA in February 2023, helping catalyze the modern open-weight model ecosystem. Llama 2 added commercial use and chat tuning, while the Llama 3 series improved scale, multilingual support, and instruction following. Llama 4, released in April 2025, introduced native multimodality and a mixture-of-experts architecture. Meta subsequently shifted its actively promoted assistant-model line to the proprietary Muse family, but Llama models remain widely used and deployed.

Initial release: 2023-02-24

Current generation: Llama 4

Reference

https://www.llama.com/

Further Reading

• https://arxiv.org/abs/2302.13971
• https://ai.meta.com/blog/llama-4-multimodal-intelligence/