GPT-6 Sol and Luna push the cost efficiency frontier by halving token cost (artificialanalysis.ai)

OpenAI has unveiled GPT-6 Sol and Luna, models that significantly improve cost efficiency, halving operational expenses compared to their predecessor, GPT-5.6. With pricing now at $2/$10 per million input/output tokens for Sol and $0.10/$0.50 for Luna, users can expect to pay $1.06 and $0.07 per task, respectively. Despite these advancements, the intelligence and coding agent index scores show mixed results; Sol improves its coding capabilities, while Luna experiences a decline. Additionally, both models exhibit reduced hallucination rates, indicating enhanced reliability as they attempt fewer answers, thus lowering their wrong answer rates. These updates are crucial for the AI/ML community, as they not only make advanced AI more accessible due to lower costs but also showcase ongoing developments in model accuracy and reliability. While improvements in areas like AutomationBench-AA suggest progress in task efficiency, regressions in knowledge work evaluations like GDPval-AA v2.1 underline the challenges still faced. As AI continues to evolve, the balance between cost efficiency and performance will shape future model development and user adoption.