OpenAI's GPT-6 Luna Reaches Agent Arena's Cost-Performance Frontier at $0.05 Per Task

OpenAI's newly released GPT-6 Luna model has joined the Agent Arena Pareto frontier, achieving a net performance improvement of +1.59% at a median cost of just $0.05 per agentic task—positioning itself as the most cost-efficient model on the efficiency curve. Introduced on September 22, 2026, alongside GPT-6 Sol, Luna costs 29% less than the open-source DeepSeek-V4.1-Flash, which achieves +4.87% improvement at $0.07 per task, and 76% less than the frontier-leading open model Hy4 preview, which delivers +5.4% improvement at $0.22 per median task.