Three weeks ago, OpenAI launched Astra, a flagship with record benchmarks (97.6% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, 100% on ExploitBench). Sol and Luna bring the same improvements in reasoning, factual accuracy, and code handling to more affordable and faster models.
Sol is designed for complex tasks like programming, with OpenAI claiming "Astra-level reliability" and half the factual errors compared to GPT-5.6 Sol. Luna is aimed at mass-market simple tasks: summarization, answering questions, and document handling.
Prices have halved: Sol is now $2/$10 per million tokens for input/output (was $4/$20), Luna is $0.10/$0.50 (was $0.20/$1.20). OpenAI attributes this to improvements in caching and inference.
The models are already available in ChatGPT Work, Codex for paid accounts, and via API, with Luna also rolling out on Free/Go.
A funny timing coincidence: Anthropic released Opus 5.5 just an hour and a half before this announcement, and OpenAI compares Sol specifically with Anthropic's models, Fable and Opus, claiming superiority. The race is happening in real-time.
https://openai.com/index/introducing-gpt-6-sol-and-luna/
#openai@rvnikita_blog #gpt6@rvnikita_blog #ai@rvnikita_blog #llm@rvnikita_blog
