OpenAI Launches GPT-5.6 Series With Three Models; Sol Improves Efficiency by 54%, Terra Cuts Pricing in Half

OpenAI on Thursday (July 9) released its GPT-5.6 model series with three versions named after celestial bodies: the flagship GPT-5.6 Sol, balanced GPT-5.6 Terra, and lightweight GPT-5.6 Luna. According to CEO Sam Altman, Sol improves token usage efficiency by 54% compared to its predecessor in AI agent programming tasks, with performance matching or exceeding top competitors.

Third-party evaluator Artificial Analysis ranked Sol first globally in coding ability with 80 points, surpassing Anthropic's Claude Fable 5 at 77.2 points, while using less than half the output tokens and reducing costs by approximately one-third. GPT-5.6 Terra matches the previous-generation GPT-5.5 but cuts API pricing in half for general workloads, while Luna offers the lowest cost with focus on low-latency, high-frequency calls. OpenAI stated Terra and Luna can achieve comparable or better outputs at approximately one-sixteenth the cost of competing products.