OpenAI’s GPT-5.6 family, released July 9, 2026, offers three tiers: Sol (flagship), Terra (balanced), and Luna (budget). Understanding their differences is essential for optimizing cost and capability.
The Three Tiers
Sol at $5/$30 per million tokens is the full flagship — maximum capability, 1M context, best benchmarks. Terra at $2.5/$15 offers 90% of Sol’s performance for half the price. Luna at $1/$6 is ideal for high-volume, cost-sensitive workloads where absolute quality is less critical.
Benchmark Comparison
On SWE-bench, Sol scores 64.6%, Terra approximately 58%, and Luna approximately 45%. On Chatbot Arena, Sol ranks ~1486, Terra ~1430, and Luna ~1350. The gap between tiers is consistent — each step down saves roughly 50% cost for 10-15% performance reduction.
Which to Choose
Use Sol for production systems where quality matters most. Use Terra for development and normal workloads. Use Luna for bulk processing, classification, and cost-sensitive pipelines. The family approach lets enterprises optimize spend across use cases.
Related: AI Model Comparison 2026
