GPT-6 Astra beats GPT-5.6 on SVG image quality in hands-on test

single source · 1 articles · capability · confidence: medium · first seen 2026-09-04 23:59 UTC

Simon Willison compared OpenAI's GPT-6 Astra with GPT-5.6 Sol, Terra and Luna by generating SVG images of pelicans riding bicycles at five reasoning levels. Astra, which does not support reasoning=none, produced better images at every level from low to max, with every Astra output rated above the best GPT-5.6 Sol image. Its list price is roughly twice Sol's — $10 per million input tokens and $50 per million output, against $5 and $30 — but it used fewer tokens, so a low-reasoning Astra image cost 9.55 cents, less than any comparable result from other models. Astra and Luna both used 16 input tokens; Sol and Terra used 26.

What this means for you

If you use OpenAI models for SVG generation, this test suggests GPT-6 Astra is worth trying: it produced better images than GPT-5.6 Sol at comparable or lower token cost per image. But it is one user's informal comparison, not a formal benchmark.

Key facts

  • ·Simon Willison compared GPT-6 Astra with GPT-5.6 Sol, Terra and Luna on SVG pelican images. source
  • ·Astra does not support reasoning=none. source
  • ·Astra list price is $10 per million input tokens and $50 per million output tokens; Sol is $5 and $30. source
  • ·A low-reasoning Astra SVG image cost 9.55 cents. source
  • ·Astra and Luna both used 16 input tokens; Sol and Terra used 26. source
  • ·Every Astra image from low to max was rated above the best GPT-5.6 Sol image. source

What the sources say

Sources

The original reporting. Follow these — they did the work.

← the wire