static.simonwillison.net web signal

Willison benchmarks GPT-6 Astra pelicans vs GPT-5.6 tiers

By Published Updated

TL;DR

  • Simon Willison ran his pelican-on-a-bicycle SVG test on GPT-6 Astra alongside GPT-5.6 Sol, Terra and Luna at reasoning-effort levels from low up to max.
  • At max effort, one Astra pelican render used 12,638 output tokens for 63.21 cents; the same prompt on GPT-5.6 Luna cost 1.57 cents.
  • Listed per-million-token prices sit next to the grid: Astra $10/$50, Sol $4/$20, Terra $2/$12, Luna $0.20/$1.20, with xhigh roughly halving both tokens and cost.

Simon Willison put OpenAI's newly available GPT-6 Astra through his pelican-riding-a-bicycle SVG benchmark and dropped the results into a side-by-side grid against GPT-5.6 Sol, Terra and Luna at reasoning-effort levels from low all the way to max.

"Got access to GPT-6 Astra. Want to see some pelicans? Yeah you want to see some pelicans," Willison wrote on X, pointing readers to the grid.

The billing table is where the comparison earns its keep. At max effort, one Astra render came out to 63.21 cents on 12,638 output tokens. The same prompt on GPT-5.6 Sol used 16,180 output tokens for 32.37 cents; Terra took 21,390 for 25.67 cents; Luna needed 13,040 for 1.57 cents. Dropping from max to xhigh roughly halved token counts and price across every tier.

Per-million list prices printed alongside the pelicans: Astra $10 in, $50 out; Sol $4/$20; Terra $2/$12; Luna $0.20/$1.20. Three experts in our Who's Who shared the grid the same day.

Shared on Bluesky by 3 AI experts

Originally reported by static.simonwillison.net

Read the original article →

Original headline: GPT-6 and GPT-5.6 SVG reasoning-effort comparison