Prompt Cache Retention: Cut OpenAI Bills Up to 90%

On July 9, 2026 OpenAI shipped GPT-5.6 to general availability as a three-tier family: gpt-5.6-sol ($5 / $30 per million input/output tokens), gpt-5.6-terra ($2.50 / $15), and gpt-5.6-luna ($1 / $6). The bare alias gpt-5.6 routes to Sol. Access is self-serve, so a lot of teams flipped their production traffic over the same week and then watched the bill climb.

Here is the part nobody puts on the launch slide: for most agentic and chat workloads, the single biggest lever on that bill is not the model tier you pick. It is prompt caching. If your system prompt, tool list, and few-shot examples repeat across calls (they almost always do), OpenAI can reuse the cached prefix and charge you roughly 90% less for those input tokens, while also cutting latency by up to 80%.