pricing inference-optimization gpu-kernels model-selection agent-deployment

OpenAI cuts GPT-5.6 Luna inference costs eighty percent

GPT-5.6 Sol autonomously rewrote production GPU kernels in Triton and Gluon, reducing end-to-end serving costs 20% and enabling Luna to undercut Gemini Flash-Lite and Claude Haiku on price.

Summary

Luna's new $0.20/$1.20 per-million-token pricing inverts model selection economics for cost-sensitive agents and background tasks. Developers running Gemini or Claude can now migrate to cheaper inference without rewriting application logic.

Why it matters

Implementation verdict

Drop-in replacement for Gemini Flash-Lite and Claude Haiku in agentic workflows. Requires testing Luna's reasoning quality against your specific use case—price alone doesn't guarantee equivalent output. Worth migrating existing agents now if latency isn't critical.

Sources

1.With Codex, GPT‑5.6 Sol autonomously rewrote and optimized our production kernels, the core code that executes the mathematical operations that make up the model.
2.we've trained GPT‑5.6 to be effective at writing and improving kernels in Triton⁠and Gluon⁠, two open-source GPU programming languages maintained by OpenAI
3.These efforts, combined with broader kernel advancements from GPT‑5.6 Sol, reduced end-to-end serving costs by 20%.
4.At $0.20/million tokens for input and $1.20/million for output Luna is now cheaper than Google's Gemini 3.1 Flash-Lite ($.025/$1.50).
5.Anthropic's cheapest current model is Claude Haiku 4.5, and that's $1/$5 - Luna is now 1/5th of that for input

Dev Signal

Get briefs like this in your inbox — free, every weekday.

100+ sources compressed into one 4-minute read. Ranked, cited, implementation-ready.

Read the full issue →All briefs