🛠️Simon Willison: Opus 5.5 vs GPT-6 Sol Sparks LLM Price War
TL;DR
Simon Willison compares Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna hours after launch, with a full price table and hands-on tests. His sharpest finding: Opus 5.5 on max effort burned all 128,000 output tokens reasoning about an SVG and returned nothing, twice.
Simon Willison compares Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna hours after launch, with a full price table and hands-on tests. His sharpest finding: Opus 5.5 on max effort burned all 128,000 output tokens reasoning about an SVG and returned nothing, twice.
Key Points
Side-by-side table puts GPT-6 Luna at $0.10/$0.50 and Grok 4.7 at $2/$6 per million tokens
Two failed Opus 5.5 max runs cost $2.56 each and took nearly 20 minutes
Opus cache-read price fell 60%, which matters most for long agent sessions
GPT-5.6 has a 25% price hike scheduled for November, widening the gap
Willison now defaults to GPT-6 Sol in Codex and Opus 5.5 in Claude Code
Why It Matters
The practical takeaway: pick effort levels deliberately, because max reasoning can overthink simple jobs and cost real money with nothing to show for it.
Quick Facts
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.