Chat GPT: OpenAI Reveals GPT-5.6 Sun, Earth and Moon: The New Era of Artificial Intelligence
- Eduardo Costa
- Jun 29
- 3 min read
Updated: Jul 2
OpenAI has once again shaken up the tech world by announcing the limited preview release of its latest and most advanced line of Artificial Intelligence models: the GPT-5.6 series .
Comprising three distinct models — Sol , Terra , and Luna — this new generation promises to redefine what we expect from generative AI, with a massive focus on agentic capabilities, advanced cybersecurity, and greater cost efficiency.
If you want to know what changes with GPT-5.6 Sol and how this update will impact the technology market and software development, keep reading.
What is the GPT-5.6 chat series? Learn about the three models.
OpenAI has adopted a new nomenclature. Now, the number identifies the model generation, while the names represent the capability levels ( tiers ):
GPT-5.6 Sol: The top-of-the-range ( flagship ) model. It is the most intelligent, powerful, and robust system created by OpenAI, designed for extremely complex tasks.
GPT-5.6 Terra: The balanced model for everyday work. It offers highly competitive performance (comparable to the previous GPT-5.5), but at twice the price .
GPT-5.6 Luna: The ultra-fast and economical option, bringing strong logic capabilities at the lowest cost on the market.
The Big News: "Max" and "Ultra" Modes
The major architectural revolution in GPT-5.6 Sol lies in how it processes deep thought. OpenAI has introduced two new reasoning functionalities:
"Max" Reasoning Effort : Allows the Sol model to take as much time as necessary to reason deeply about highly complex problems before giving an answer.
"Ultra" mode: Goes far beyond the capabilities of a typical virtual assistant. Ultra mode allows the main agent to recruit and coordinate AI sub-agents to accelerate and automate massive workflows.
Performance in Benchmarks: Absolute Leadership
Data shared by OpenAI shows that GPT-5.6 Sol has set new industry records ( State of the Art ):
Programming and Code: In Terminal-Bench 2.1 (which tests command-line workflows that require planning and coordination of tools), GPT-5.6
Sol Ultra achieved an impressive 91.9% score, far surpassing competitors such as the Claude Mythos 5 and the Gemini 3.1 Pro.
Quantitative Biology: In GeneBench v1 , focused on long-term genomic analyses, the Sol model outperformed GPT-5.5 while using significantly fewer tokens (which translates to lower latency and cost).
Cybersecurity: On ExploitBench , Sol proved to be highly competitive, using only one-third of the tokens of rival models to identify vulnerabilities and suggest fixes ( patches ).
Layered Security: Automated Red-Teaming
Because it is a model with dangerously advanced cyber capabilities, OpenAI has implemented its most robust security stack to date.
The model underwent over 700,000 hours of computing (equivalent to A100 GPUs) in automated Red-Teaming . This means that OpenAI used its own AI systems to continuously attack GPT-5.6, attempting to find universal jailbreaks (security flaws) and shield the system against real-world attacks before release.
Additionally, the system has real-time classifiers. If it detects a suspicious or dangerous request in the area of biology or cybersecurity, generation is paused and a superior reasoning model evaluates the context to decide whether to block the response.
Availability: When can you use it?
For now, access to GPT-5.6 Sol is restricted to a very select group of trusted partners and organizations, in a process closely coordinated with the US government due to the model's strong cybersecurity capabilities.
However, OpenAI has already confirmed that it plans to make GPT-5.6 Sol, Terra and Luna generally available to all ChatGPT, Codex and API users in the coming weeks .
Keep an eye on our blog for more updates and tutorials as soon as this tool becomes available to the general public!
Comments