The Pelican comparison grid for Astra is pretty interesting
After obtaining access to GPT-6 Astra in the afternoon, users generated SVG images of cycling penguins at different reasoning levels (low, medium, high, xhigh, max) and compared them with models GPT-5.6 Sol, Terra, and Luna using a grid display. The results showed that the penguins generated by Astra had significantly better quality than those produced by other models. Specifically, Astra max performed extremely well, while at lower levels there were occasional errors in leg drawing, but overall it still outperformed all levels of Sol. In terms of cost, the unit price of Astra was about twice that of Sol (input $10/million, output $50/million, compared to Sol’s $5/$30), but due to significantly less token consumption, the actual price difference between the levels became smaller. Specifically, the low-level Astra, costing $9.55, produced results better than any level of Sol models, while using other models at $10 yielded poorer results. In terms of the number of tokens used, Astra…