GPT-5.6 Terra Has No Win Condition: Luna and Sol Dominate It on Both Price and Intelligence
Artificial Analysis published its intelligence-cost efficiency breakdown of the GPT-5.6 family, and the conclusion on Terra is unambiguous: it has no Pareto-optimal position. At every reasoning effort level, buyers can find a Luna or Sol configuration that is either more intelligent for the same cost or equally intelligent at lower cost.
What the Curve Shows
The analysis plots GPT-5.6 Sol, Luna, and Terra across reasoning effort levels on an Intelligence Index vs. Cost per Intelligence Task chart. Sol and Luna together form the efficiency frontier. Terra sits below it at every point.
The practical implication is straightforward. If a workload justifies Terra’s pricing, there is a Sol or Luna setting that beats it on intelligence at that price point. If a workload needs Terra’s intelligence level, Luna reaches it at lower cost. There is no combination of budget and intelligence requirements where Terra is the rational selection.
Luna as the Cost-Efficient Default
Luna stands out as the most cost-efficient option in the GPT-5.6 family. It reaches competitive intelligence scores — per the Artificial Analysis index — at a fraction of the cost of higher-effort Sol and Terra settings. For enterprises processing large volumes of agentic tasks, Luna offers more head-room on price-per-task than either of its siblings.
Sol is the peak-intelligence option. For workloads where the ceiling matters — frontier coding agents, complex reasoning pipelines, multi-step research tasks — Sol’s higher-effort modes deliver, and they beat Terra on cost anyway.
LiveBench’s July 2026 data corroborates the pattern: on agentic coding specifically, Terra’s max-effort score is 68.0 versus Sol’s 65.6. Terra wins that one dimension by 2.4 points, but at a price premium Artificial Analysis shows is never recovered. For general-purpose intelligence, Sol leads at 82.4 to Terra’s 79.8 overall.
The Take
OpenAI shipped GPT-5.6 as a three-tier family — Sol, Terra, Luna — and Terra’s positioning is the weakest of the three. Sol is the headline model; Luna is the value play; Terra is caught between them with no clear buyer. Whether OpenAI rationalises the lineup or adjusts Terra’s pricing is the open question. For now, the guidance from the data is simple: Sol for capability, Luna for volume, Terra for neither.