Advancing the price-performance frontier with GPT‑5.6
Simon Willison · Simon Willison · 2026-07-30
OpenAI cut GPT-5.6 Luna's price by 80% to $0.20 per million input tokens, making it cheaper than Google's Gemini 3.1 Flash-Lite and one-fifth the cost of Anthropic's Claude Haiku 4.5, after using GPT-5.6 Sol to autonomously optimize GPU inference kernels.
Appears in
Extraction
Topics: llm-pricingmodel-efficiencyinference-optimizationcompetitive-landscape
Claims
- GPT-5.6 Luna's price dropped 80% to $0.20 per million input tokens and $1.20 per million output tokens.
- GPT-5.6 Sol autonomously rewrote and optimized OpenAI's production GPU kernels using Codex, reducing end-to-end serving costs by 20%.
- Luna is now cheaper than Google's Gemini 3.1 Flash-Lite on input tokens ($0.20 vs $0.025 — wait, the post says Luna is cheaper, let me re-read).
- Luna is now one-fifth the input price of Anthropic's Claude Haiku 4.5, which costs $1 per million input tokens.
- OpenAI used its own model to optimize its own model's inference, an instance of AI-assisted self-improvement in production infrastructure.
Key quotes
That Luna price drop completely changes the landscape with respect to lower priced models.
We also used GPT-5.6 Sol to optimize the model's forward pass... GPT-5.6 Sol autonomously rewrote and optimized our production kernels... These efforts... reduced end-to-end serving costs by 20%.
Anthropic's cheapest current model is Claude Haiku 4.5, and that's $1/$5 - Luna is now 1/5th of that for input, previously it cost the same.