Simon Willison finds a 68x cost spread across OpenAI's new GPT-5.6 tiers
Simon Willison benchmarked all three GPT-5.6 tiers across every reasoning level and found a 68x cost spread — Luna at zero reasoning runs 0.71 cents against Sol at max reasoning at 48.55 cents. The spread means per-million-token pricing no longer predicts real cost; reasoning effort is now the dominant cost lever inside a single model family. Willison also noted OpenAI called ~30% of SWE-Bench Pro tasks broken the day before Sol's 64.6% trailed Fable 5's 80%, and rated Sol no better than Fable at complex coding.
Source: simonwillison.net ↗
the least expensive was gpt-5.6-luna at effort none for 0.71 cents, the most expensive was gpt-5.6-sol at max reasoning level for 48.55 cents
Simon Willison
Why this matters
- → Reasoning effort now dominates pricing, not model size
- → 68x cost variance makes per-token pricing unreliable
- → Sol underperforms Fable on coding despite higher cost
Reasoning reshapes LLM pricing