OpenAI released two new models, GPT-6 Sol and GPT-6 Luna, on September 22, just 19 days after its flagship GPT-6 Astra. The company is positioning them as workhorses for everyday enterprise tasks that do not need Astra's top-tier reasoning.
GPT-6 Sol now costs $2/$10 per million input/output tokens, down from $4/$20 for GPT-5.6 Sol, matching Claude Sonnet 5 pricing. GPT-6 Luna drops to $0.10/$0.50 per million tokens. An OpenAI spokesperson confirmed the cuts are permanent list prices, not a limited promotion, made possible by inference and caching improvements alongside a 90% discount on cached input reads.
OpenAI says GPT-6 Sol makes roughly half as many factual errors as its predecessor and scores better on agent-safety metrics such as deception and failure disclosure. Sol targets recurring complex work like code review and debugging, while Luna is aimed at high-volume summarization and extraction. Both are live in the API now and rolling out to ChatGPT Work and Codex subscribers, with Luna also reaching free users through the desktop app.