Advancing the price-performance frontier with GPT‑5.6

OpenAI has introduced significant price reductions for its GPT-5.6 models, cutting prices by 20 percent for GPT-5.6 Terra and by 80 percent for GPT-5.6 Luna. These cost savings were driven by another model, GPT-5.6 Sol, which was used to improve load balancing and optimize the core inference process.

By leveraging Codex along with Triton and Gluon programming languages, GPT-5.6 Sol autonomously rewrote and optimized production kernels. This eliminated excess memory movement, reduced synchronization delays, and better parallelized GPU workloads, ultimately cutting overall serving costs by 20 percent.

These technical improvements allow the Luna model to be offered at $0.20 per million input tokens and $1.20 per million output tokens. This new pricing structure makes Luna substantially cheaper than competing options, undercutting Google's Gemini 3.1 Flash-Lite and costing one-fifth of Anthropic's Claude Haiku 4.5.