- /
- Creators/
- Simon Willison — Creator of Datasette; co-creator of Django/
- Advancing the price-performance frontier with GPT‑5.6/
Advancing the price-performance frontier with GPT‑5.6
Blog · Read
- OpenAI cut GPT-5.6 Terra pricing 20% and GPT-5.6 Luna pricing 80%; Luna now runs $0.20/M input, $1.20/M output — a fifth the cost of Claude Haiku 4.5 ($1/$5), positioned against Google’s Gemini 3.1 Flash-Lite ($0.025/$1.50).
- OpenAI credits its internal model GPT-5.6 Sol for the infrastructure optimization behind the cut — used it to rewrite inference kernels in Triton and Gluon (GPU programming languages), cutting serving costs 20% by eliminating GPU idle time.
- Willison switched his agent.datasette.io demo from Gemini to Luna as a direct result of the price change.
- Takeaway: frontier labs are now using their own models to optimize the infrastructure that serves them, compounding cost reductions beyond training-side gains alone.