The Pelican comparison grid for Astra is pretty interesting

A model's price per unit of generated text does not by itself determine the cost of finishing a task. Simon Willison generated SVG drawings of a pelican riding a bicycle with OpenAI's GPT-6 Astra at several reasoning levels and compared them with GPT-5.6 Sol, Terra and Luna in a comparison grid. Astra's listed prices ($10 per million input tokens and $50 per million output) are roughly twice Sol's ($5 and $30), but Willison reports that Astra used significantly fewer tokens, the text units an API bills for, at each reasoning level, making the prices at the different levels closer than they might otherwise be.

The useful observation is about total work consumed: a higher rate can be partly offset by a shorter generation. This was one drawing prompt, and Willison's quality judgments are personal, so it offers a task-level comparison rather than a general cost or quality ranking.