OpenAI slashed pricing for two models in its GPT-5.6 lineup on July 30, cutting Luna—the smallest, fastest tier—by 80% to $0.20/$1.20 per million input/output tokens, and Terra by 20% to $2/$12, while leaving flagship Sol unchanged. The cuts arrived just three weeks after the GPT-5.6 family launched and reflect system-wide efficiency gains: GPT-5.6 Sol autonomously rewrote its own production GPU kernels and speculative-decoding models, reducing serving costs by 20–15% through self-optimization.
Luna now offers ~87% lower cost than GPT-5.4 mini for the same reasoning ability—a 13x reduction in four months—and sits below Google's Gemini 3.6 Flash ($1.50/$7.50). The move undercuts Chinese rivals DeepSeek V4 Pro on input costs and targets enterprise buyers who'd shifted to cheaper models. Anthropic's Claude Opus 5 at $5/$25 remains more capable but costlier; Google, Microsoft, and Anthropic all cut prices weeks earlier, triggering the race.
Architects should note the dual-tier strategy: Luna commoditizes inference for high-volume workflows (agents, classification, summarization), while a new Sol Fast mode charges 2× standard price for up to 2.5× throughput without intelligence change—monetizing latency separately. OpenAI signaled this is structural, not temporary: CEO Sam Altman said costs were "a huge issue" and the company would pass efficiency gains to users. For production systems, Luna's cost basis now rivals open-weight models while retaining frontier-level reasoning on common benchmarks.