GPT-6 Sol and Luna. OpenAI is not slowing down with its premieres

GPT-6 Sol is a mid-range model intended for more complex tasks, such as programming. Luna is a lighter and cheapest option for simple, large-scale tasks such as summarizing documents, searching for information, or quickly answering questions.

Both models do not dethrone the flagship model from OpenAI – GPT-6 Astra. The new models are expected to operate almost as reliably as Astra and cost much less. This is an approach, a bit like Anthropica, that OpenAI is now starting to adopt.

According to the company, the Sol model makes half as many errors in fact-checking tests as its predecessor.

Both models are also intended to respond shorter and more precisely.

Rates have halved compared to GPT-5.6

GPT-6 Sol costs $2 in the API. for one million input tokens and $10. for a million starting. That’s half the price of GPT-5.6 Sol, which cost $4 and $20. Luna costs $0.10. for entry and $0.50. per exit, compared to $0.20 and $1.20. before.

Test results

In Zapier’s AutomationBench test, GPT-6 Sol achieved 33.2 percent and Claude Opus 5 26.9 percent. In Sol’s case, the task cost many times less.

In DeepSWE 1.1, the code development test, Sol scored 68.8 percent and Luna scored 66.6 percent.

In OSWorld 2.0, which measures computer skills, Sol scored 60.5 percent and Opus 5 60.3 percent.

However, OpenAI compared its capabilities with the previous generation of Anthropic models, not with the latest Opus 5.5.

Where new models from Open AI are available

The models entered the API as gpt-6-sol and gpt-6-luna. They are also available in ChatGPT Work and Codex in the paid Plus, Pro, Business and Enterprise plans. Luna can also be used by users of the free version and the Go plan in the desktop application. ChatGPT will receive new models gradually.

Sources: Antyweb, TechCrunch, VentureBeat, Android Headlines