OpenAI has released GPT-5.6, a family of three models the company describes as its most capable, fastest and most cost efficient frontier system to date. The lineup consists of Sol, the flagship, alongside Terra and Luna, which are positioned for different workloads and budgets. The models are available in ChatGPT, Codex and the OpenAI API, with the global rollout completing over about 24 hours from the July 9 announcement.
Efficiency as the headline
According to OpenAI, GPT-5.6 Sol sets a new level of performance in programming, science, cybersecurity, research, business analysis and complex reasoning. The company says the model produces more accurate results while consuming fewer tokens than its predecessors, which translates into lower cost for the same work.
OpenAI frames the release around useful work per token rather than raw intelligence alone. It claims GPT-5.6 beats both its own earlier models and several rival frontier models on key benchmarks while keeping total cost substantially lower. These are company reported results and have not yet been independently confirmed.
Ultra mode runs four agents at once
A new capability called Ultra departs from the usual single model approach. Instead of one process working through a problem, Ultra dispatches four AI agents in parallel. Each works on a separate part of the task, the agents review one another’s output and the system assembles a single coherent answer. OpenAI says the approach improves both speed and quality on long, complex and research heavy projects.
Coding, design and knowledge work
OpenAI calls GPT-5.6 its most powerful coding model. Beyond writing code, the company says it can run programs, use tools, inspect intermediate results and change strategy when needed, enabling workflows that previously required multiple separate model calls or human oversight.
The model can also look at a rendered interface, identify visual flaws and improve them, which OpenAI says makes it more effective at producing websites, presentations, documents and spreadsheets. For knowledge work, GPT-5.6 has been tuned to turn scattered information from sources such as Slack, Notion, Microsoft 365 and Google Drive into organized, shareable reports and presentations, while respecting existing templates, design systems and formatting.
Cybersecurity and science
OpenAI describes GPT-5.6 as its strongest defensive cybersecurity model, with improved performance in vulnerability discovery, secure code review, patch generation and threat modeling. The company says safeguards were strengthened in parallel so that legitimate research and defensive work can proceed while malicious use is blocked.
In the life sciences, OpenAI reports gains over GPT-5.5 on benchmarks covering biology, genomics and chemistry. It also says its own researchers are using GPT-5.6 to design experiments, debug and improve training systems, and that internal research velocity has risen as a result.
Safety testing
OpenAI says GPT-5.6 ships with its most robust safety stack so far. Before general availability the model went through millions of automated tests, red team exercises by external experts and security evaluation amounting to roughly 700,000 A100 equivalent GPU hours, according to the company. The system does not rely on classifiers alone; it adds a real time reasoning monitor that weighs the context of each request before deciding whether a response is safe to give.
Availability
Plus, Pro, Business and Enterprise subscribers receive access to Sol, Terra and Luna at different tiers, and developers can call the models through the API.
The release underlines where competition among AI labs is heading. The contest is no longer only about building the smartest model, but about systems that do more work for less money, manage complex projects autonomously, coordinate multiple agents and serve as practical assistants across research, business, science and security. GPT-5.6 is OpenAI’s most ambitious step in that direction so far.
