OpenAI has unveiled the first details of its new generation of language models, GPT-5.6. Instead of a single universal model, this time there are three of them variant with a different focus. The new models are named Sol, EARTH a Moon and are intended to cover the needs of maxfrom maximum performance to the lowest possible operating costs.
This is a preview available for now. only to selected partners, wider deployment in ChatGPT but other services will arrive in the coming weeks and months.
You might be interested in
Sol will be OpenAI's new flagship
The most powerful model in the entire series is GPT-5.6 Sol. OpenAI has optimized it primarily for demanding tasks in the fields of programming, cybersecurity, and life sciences. Compared to GPT-5.5, it promises higher accuracy, better efficiency, and lower token consumption when processing complex tasks.
The model also includes new modes Max a UltraThe former allows for deeper thinking about complex problems, while the latter uses the collaboration of multiple agents to solve complicated tasks together. It is this approach that suggests where the development of modern AI systems will go in the coming years.
Terra aims for the ideal price-performance ratio
For most companies and developers, the Terra model will likely be the most interesting. It represents a compromise between performance, speed, and operating costs. OpenAI says it will offer a similar level of capabilities to today's GPT-5.5, but at less than half the price.
This can significantly reduce costs for companies that use AI on a large scale, for example for customer support, document automation or software development. Operational costs are one of the biggest issues in the current generative AI market.
Gallery
Luna builds on maximal efficiency
The third model is GPT-5.6 Luna. It is designed primarily for scenarios where price and speed are key. While it won't offer the same performance as Sol, OpenAI claims it retains very solid capabilities at a price that is more than half that of the Terra model.
Such a model may be ideal for chatbots with a high number of users, automatic translations, text summarization or other tasks where efficiency is of primary importance.
Security remains a major issue
Security is also a big focus this time around. OpenAI says all three models include multiple layers of protection against abuse. The most powerful, Sol, was specifically designed to detect security flaws in software and suggest fixes, but it refused to create complete walkthroughs or exploits that attackers could directly use.
The company also admits that in the first testing phase, the protection mechanisms may be too strict. In some cases, they may block legitimate user requests. Therefore, OpenAI expects to further tune them based on feedback from partners.
What does this mean for users? ChatGPT?
The change in strategy shows that the development of generative AI is shifting from general-purpose models to specialized ones variaMore and more companies are choosing a similar approach today, as it allows them to better tailor performance to specific applications while significantly reducing operating costs.
For regular users ChatGPT The most important news will be that the new models will also be coming to this service. OpenAI has not yet announced an exact schedule, but after the preview program ends, they should appear not only in ChatGPT, but also in the Codex tool and other API services.
The preview is currently only available to trusted partners and organizations around the world. No regional restrictions have been announced, so it is expected to be available in the European Union at public launch. However, as with previous models, some advanced features or modes may be available over time and availability may vary by subscription or service.
