GPT-5.6 made advanced AI cheaper to repeat

OpenAI’s GPT-5.6 update moves AI competition down the cost curve. The Luna tier is now priced at $0.20 per million input tokens and $1.20 per million output tokens, after an 80% price cut; Terra is 20% cheaper.1

The cut came from several layers. OpenAI says GPT-5.6 was trained to complete more work per token; GPT-5.6 Sol helped rewrite production GPU kernels, cutting end-to-end serving costs by 20%; and its experiments raised token-generation efficiency by more than 15%.2

The agent layer changed too. OpenAI’s harness trims context bloat, preserves exact prompt prefixes for caching, and keeps tools from repeating work across multi-step tasks.2

The practical shift is a wider ladder: Sol for the hardest reasoning, Terra for balanced work, and Luna for high-volume tool use. More AI work can fit inside the same budget when each step gets a model tier matched to its cost and latency.1