OpenAI’s New GPT-6 Sol and Luna Models Are 50% Cheaper

OpenAI has added two new models to the GPT-6 family. GPT-6 Sol and GPT-6 Luna bring much of the flagship GPT-6 Astra‘s performance into faster and cheaper models designed for work at scale, with both rolling out starting today.

OpenAI says improvements to caching and inference have lowered its costs, and some of those savings are being passed on to developers. GPT-6 Sol now costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20. GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens, down from $0.20 and $1.20.

Both models are a whopping 50% cheaper than their GPT-5.6 counterparts.

OpenAI is also pitching the new models as capable enough to handle serious business and coding workloads. On AutomationBench, GPT-6 Sol at its highest effort scores 33.2%, beating Claude Opus 5 and Claude Fable 5.1 at a much lower cost, while also topping GPT-6 Astra running at low effort.

GPT-6 Luna, meanwhile, scores 66.6% on the DeepSWE coding benchmark, which OpenAI says puts it in the same range as Claude Opus 5 at medium effort. GPT-6 Sol and Luna are rolling out today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. Both are also available through the API, while Free and Go users can try GPT-6 Luna in the desktop app.

The company says Sol and Luna also build on Astra’s alignment work, which OpenAI describes as its most aligned model to date. New caching tools, including a prompt caching dashboard, are aimed at helping developers cut costs further on longer conversations and repeated workloads.

The launches come after OpenAI slowed development of a follow-up model over hacking concerns earlier this summer. Competition remains intense too, with SpaceXAI recently launching Grok 4.7.

For users and developers, the bigger story here is the cost. Cheaper models that retain much of the performance of a flagship system can make AI features easier to deploy across coding tools, business software and consumer apps without the same compute bill. Let the AI race continue, folks, as here we are with yet another announcement and cheaper pricing.