How to Choose an AI Model

With the growing number of AI models, it is not easy to find the right one for your use case. Two websites are particularly helpful: Artificial Analysis and the LM Arena Leaderboard.

Artificial Analysis

Artificial Analysis provides independent benchmarks for intelligence, speed, and cost across models. The site computes its own Intelligence Index, combining nine evaluations — from agentic coding (Terminal-Bench) to scientific reasoning (GPQA Diamond) to knowledge tests (AA-Omniscience).

The big advantage: you can compare models not only by quality but also by cost per task. This is especially relevant when deploying a model in production. A model that delivers 95% of the top performer’s quality at a tenth of the cost is often the better choice.

There is also a dedicated Coding Agent Index that evaluates end-to-end software engineering tasks.

LM Arena Leaderboard

LM Arena takes a different approach: real users rate responses in blind tests, and an Elo ranking is derived from their preferences — similar to chess ratings.

The leaderboard is broken down by category: Coding, Math, Creative Writing, Instruction Following, and more. This lets you see at a glance which model leads in your target domain.

Which Model Should I Choose?

The best approach is to combine both sources:

1. Define your use case: Do I need a model for code, creative writing, or general reasoning?
2. Check LM Arena: Which models perform well in the relevant category?
3. Consult Artificial Analysis: How do those models compare in terms of speed and cost?
4. Consider your budget: Open-weight models (e.g. GLM-5.2, DeepSeek, Qwen) can be self-hosted. Proprietary models (Claude, GPT, Gemini) often offer the highest quality but charge per API call.

A quick rule of thumb: for the best quality, look at the LM Arena top 10. For the best value for money, filter by cost per task on Artificial Analysis.