OpenAI has unveiled its next version of its artificial intelligence model, namely GPT-5.6 family, which consists of three different models: GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna. OpenAI claims that GPT-5.6 family offers enhanced abilities in terms of reasoning, coding, cybersecurity and scientific research.
However, contrary to previous generations of the model, the introduction of this new family of AI models will be started with a preview mode, which is going to be available only for selected partners in the United States.
The unveiling of this new family of models takes place against the background of increasing attention to AI technologies from regulatory bodies. According to the company, the US government was informed about the potential of new models. This is due to the current development of evaluation frameworks for frontier AI systems in accordance with cybersecurity regulation.
Three models designed for different tasks
OpenAI has adopted a new naming structure for the GPT-5.6 family, with the numerical designation representing the model generation and the names Sol, Terra and Luna indicating different performance tiers.
Also Read | India climbs to 13th globally in QS Future Skills Index 2027, signalling a stronger AI-ready workforce
GPT-5.6 Sol has been positioned as the company's flagship model for advanced reasoning, scientific research and complex coding tasks. GPT-5.6 Terra targets enterprise and developer workloads while offering performance comparable to GPT-5.5 at approximately half the cost. GPT-5.6 Luna has been designed as the fastest and most cost-effective option for applications where affordability and speed are priorities.
GPT‑5.6 Sol launches with our most robust safety stack yet.
— OpenAI (@OpenAI) June 26, 2026
We strengthened real-time protections against high-risk cyber activity and repeated misuse, then spent weeks hardening the system with human red teaming and over 700,000 A100-equivalent GPU hours of automated testing.
The company also announced that GPT-5.6 Sol will run on Cerebras hardware beginning in July, enabling inference speeds of up to 750 tokens per second for selected customers.
Reasoning, Coding and cybersecurity see major upgrades
According to OpenAI, GPT-5.6 introduces a new "Max Reasoning Effort" feature that allows the flagship Sol model to spend more time analysing difficult problems before generating responses.
Another major addition is "Ultra Mode", which deploys multiple specialised sub-agents to solve complex tasks collaboratively rather than relying on a single AI system. OpenAI says the feature improves performance across coding, research and multi-step workflows.
The company reported that GPT-5.6 Sol delivered stronger performance in command-line coding benchmarks, planning tasks and computational biology evaluations while requiring fewer output tokens than previous models.
Cybersecurity capabilities have also received substantial upgrades. OpenAI said GPT-5.6 Sol demonstrated improved performance in vulnerability research and defensive security testing, matching leading benchmark performances while using significantly fewer computational resources.
However, OpenAI stated that the model remains below its internally defined "Cyber Critical" threshold and was unable to autonomously execute complete end-to-end cyberattacks during internal evaluations.
Expanded safety framework introduced
OpenAI said GPT-5.6 launches with its most comprehensive safety architecture to date. The framework includes models trained to refuse prohibited cyber assistance, real-time monitoring systems, account-level misuse detection and extensive automated and human-led red-team testing.
Also Read | Alibaba denies nothing yet as Anthropic alleges massive Claude AI extraction campaign
According to the company, it invested more than 700,000 A100-equivalent GPU hours in automated safety assessments, in addition to the weeks of manual testing aimed at finding vulnerabilities and misuse cases.
In explaining the limited access to GPT-5.6, OpenAI emphasized that it does not see the preview requirements by the governments as a long-term model for other frontier AI systems and intends to expand GPT-5.6 access to ChatGPT, the API platform and Codex in the coming weeks.
Such developments highlight the conflict between advancements in artificial intelligence technologies and the need to regulate these sophisticated systems.
FAQs
Q1. Which models are in OpenAI's GPT-5.6 family?
Three models make up the GPT-5.6 family; GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna.
Q2. Why is OpenAI making GPT-5.6 access limited initially?
OpenAI said it agreed to the limited US preview of GPT-5.6 after briefing government officials, amid the development of AI assessment frameworks.
