← Back to the wire

The Pelican comparison grid for Astra is pretty interesting

AchievementBenchmarkSep 4, 2026

In a pelican-riding-bicycle SVG comparison, GPT-6 Astra at its low reasoning level produced better results than any GPT-5.6 Sol model at any level, for 9.55 cents. Astra pelicans from low to xhigh all outperformed the best Sol output. Astra costs roughly twice Sol's price but uses fewer tokens. Astra and Luna both used 16 input tokens versus 26 for Sol and Terra, prompting speculation about their relationship.

Receipt № 17451 source · awaiting confirmation ◐

Evidence

1source· awaiting independent confirmation

No score is assigned. Sources and their independence are shown in the citation chain below.

Citation chain · 1 source

LunaModelTerraModelGPT-5.6 SolModelGPT-6 AstraModelGPT-6 NovaModel
Canonical: https://simonwillison.net/2026/Sep/4/astra-pelicans/