ArtificialAnalysis calculates and publishes the intelligence score of each LLM model alongside the cost required to run benchmarks. The site's "Intelligence vs. cost" plot visualizes the Pareto optimality, showing which models are the most cost-effective for achieving specific intelligence scores.
However, an engineer has pointed out three issues regarding this plot.
First, the cost axis uses a logarithmic scale (log scale). It was stated that while this emphasizes slight price differences between inexpensive models, it risks visually underestimating the overwhelming price gaps compared to expensive models.
Second, the cost calculation is based on the API prices set by the model developers themselves. For open-weight models, it is possible to use them more cheaply by utilizing third-party providers such as OpenRouter.
Third is the handling of models that can run in local environments. It was pointed out that because these models are plotted using data center pricing, they deviate from the practical sense of cost.
For some small-scale models, execution costs based on electricity rates are calculated.
--- Source: LLMs: Intelligence vs. Cost (Hacker News Frontpage, 2026-09-02)