Three weeks after launching the GPT-5.6 lineup, OpenAI cut API pricing for its two lower tiers: GPT-5.6 Luna dropped from $1/$6 to $0.20/$1.20 per million input/output tokens (an 80% cut), and GPT-5.6 Terra dropped from $2.50/$15 to $2/$12 (a 20% cut); the flagship Sol model kept its price but gained a new "Fast mode" running roughly 2.5x faster at double the cost, replacing the old Priority Processing tier. OpenAI frames this as passing along real efficiency gains — better routing, more efficient production kernels (some of which Sol reportedly helped rewrite itself), and smarter context management that avoids agents re-processing work they've already done — rather than a simple promotional discount. The practical significance for developers building with LLM agents is bigger than a line-item cost change: several companies quoted in OpenAI's own announcement (Cognition, Dust, Replit, Blitzy) describe restructuring their production agent pipelines specifically around Luna as the new default "workhorse" model for high-volume, well-specified steps — implementing a pre-planned change, running and evaluating tests, executing scoped tool calls — while reserving Sol for the harder planning and disambiguation steps. That two-tier pattern (a strong, expensive planner model handing off well-defined subtasks to a much cheaper execution model) is becoming a standard cost-control architecture for production agents, and this price cut makes that pattern significantly more attractive economically, since the cost gap between "smart enough to plan" and "cheap enough to execute at scale" just widened further in the cheap direction. Teams running any kind of multi-step agent loop today have a direct incentive to revisit their model-routing logic in light of this.