What LLM to use today?

2025-11-28

Summary: A sourced summary about choosing AI models based on real tasks instead of only benchmark numbers.

This sourced note explains that public benchmarks can be useful but often fail to show how a model will feel in real work. The author argues that instruction following, agentic ability, and how well the model understands your intent can matter more than small benchmark differences.

The practical advice is to test several models on real tasks that matter to you. Ask: does the model understand what you want, complete the task properly, avoid annoying mistakes, and fit your workflow? Different models have different strengths, so the best model depends on the job.

Free Basics version: do not choose an AI tool only because it is popular. Try it on your actual writing, learning, coding, or planning tasks.

Source: Daniil Sedov, Gusarich's thoughts, gusarich.com/blog/what-llm-to-use-today