With new LLM releases dropping almost every week, selecting the best model for the job is not trivial. Following official announcements and cutting through marketing noise, or trying all of them would be a full time job.
If I had to pick the single most useful chart of the year, it's the Intelligence vs. Price chart on Artificial Analysis. It shows the most capable models in each price point, and captures a lot of benchmark results in a single score.
Here's a snapshot from today, with my filters applied to only show models available in GitHub Copilot subscription:
It's worth to note some aspects:
- The price scale on X-axis is logatyrhmic, not linear. Price differences are huge, with some models being 100x the cost of the cheapest ones.
- Reasoning level is a very important choice. Some models might be on the pareto frontier with a given reasoning level, while being far from the best choice at other reasoning levels (see for example Gemini 4 Hydrogen on a chart).
Sooner or later the best option might be to delegate this decision to specialized orchestrators like Copilot Fusion or Cloudflare AI Gateway's Auto Router. I see those orchestrators as quite new and slightly lagging behind the AI model releases, so for now I still like to be able to choose the most promising model for the task myself.