OpenAI introduced GPT-6 Sol and GPT-6 Luna, expanding its GPT-6 model family with options designed for varying balances of capability and cost. The new models are available through OpenAI's API, ChatGPT Work, and Codex, with Luna also reaching Free and Go users in the desktop application. This release follows the earlier launch of the flagship GPT-6 Astra model.
GPT-6 Sol and Luna are priced at half or less than the cost of their GPT-5.6 predecessors when accessed via OpenAI's API. GPT-6 Sol is priced at $2 per million input tokens and $10 per million output tokens, a 50% reduction from GPT-5.6 Sol's rates of $4 and $20, respectively. GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens, representing a 50% reduction for input and a 58.3% reduction for output compared to GPT-5.6 Luna's $0.20 and $1.20 rates. An OpenAI spokesperson confirmed these are permanent prices, not promotional offers. OpenAI attributes these price reductions to advancements in inference and caching.
The company also reported improvements in factual reliability and coding performance for the new models. In internal evaluations based on de-identified ChatGPT conversations where users had previously identified factual errors, GPT-6 Sol made approximately half as many mistakes as GPT-5.6 Sol. OpenAI stated that GPT-6 Sol is "approaching Astra-level reliability" at a lower cost. GPT-6 Luna, at higher effort levels, can match GPT-5.6 Sol's factuality at about one-hundredth of its task cost, according to OpenAI.
For coding tasks, GPT-6 Sol achieved a score of 68.8% on the DeepSWE 1.1 benchmark at maximum effort. OpenAI estimates Sol's cost per task to be roughly 80% lower than Claude Fable 5, which scored 69.9% at xhigh effort. GPT-6 Luna reached 66.6% on the same benchmark, with OpenAI claiming its task cost is 93% below a comparable Opus 5 configuration and 96% below Fable 5. On OSWorld 2.0, a computer-use benchmark, Sol xhigh scored 60.5% in OpenAI's testing, compared to 60.3% for Claude Opus 5 at medium effort, with an estimated 80% lower cost per task for Sol.
GPT-6 Sol is designed for complex work such as building features, reviewing code, debugging, and analyzing data. GPT-6 Luna is positioned as a high-volume option for more routine tasks, including fact extraction, text sorting, and structured summaries. Both models are built on the alignment work introduced with Astra, which OpenAI describes as its most aligned model to date. In alignment evaluations, both Sol and Luna showed improvements over their GPT-5.6 counterparts, including lower rates of misleading claims about their coding work. OpenAI CEO Sam Altman stated that the company aims to offer leading AI models across every price point and modality.