
Tech • IA • Crypto
OpenAI has launched its GPT‑5.6 model family alongside new productivity tools, aiming to transform AI from a chat assistant into a full-fledged work platform.
OpenAI introduced four models, with GPT‑5.6 Soul positioned as the most advanced for complex, long-running tasks. Terra targets everyday workflows, while Luna is optimized for speed and affordability in high-volume use. The rollout begins within 24 hours, with Soul for paid users and Terra and Luna also reaching free tiers.
The company reported nearly 1 billion weekly users, highlighting a shift from simple queries to advanced applications such as coding, research, and creative production. Internal usage has expanded significantly, with employees relying on AI for operational and strategic tasks rather than just information retrieval.
A new product, ChatGPT Work, is designed as a task-oriented assistant capable of handling complex workflows. It can analyze data, manage long-horizon projects, schedule meetings, and synthesize information from tools like Slack and internal feedback systems. Demonstrations showed it autonomously identifying relevant stakeholders and arranging meetings based on user goals.
In finance use cases, ChatGPT Work reduced hours of manual effort by performing variance analysis, updating forecasting models, and generating presentations automatically. It can integrate outputs across Excel, PowerPoint, and interactive dashboards, enabling teams to operate more efficiently with fewer resources.
An updated desktop application allows the AI to interact directly with local files, browser tabs, and software. It can process large datasets, generate visualizations, organize notes autonomously, and compile multi-source reports into ready-to-use presentations in under two minutes.
The models can generate real-time interactive charts, dashboards, and even 3D visualizations without manual coding. These outputs are dynamic and shareable, enabling teams to collaborate through automatically generated interfaces rather than static documents.
A new feature called Hosted Sites enables users to convert AI outputs into fully functional, shareable websites. These sites can include dashboards, reports, or prototypes, allowing teams to present and distribute work without traditional development pipelines.
GPT‑5.6 demonstrates state-of-the-art performance in coding benchmarks and agent-based tasks. Internally, Soul has autonomously executed research workflows, including training other models, signaling progress toward semi-automated AI research and engineering processes.
Despite increased capability, OpenAI emphasized token efficiency, claiming GPT‑5.6 delivers higher performance at less than half the cost of comparable systems on certain benchmarks. A new “ultra mode” allows multiple AI agents to collaborate in parallel for faster and more accurate results.
The company invested over 700,000 GPU hours in safety testing and red teaming. Early deployments with cybersecurity researchers identified vulnerabilities in major systems, while new initiatives have contributed patches to open-source projects, including Linux.
Demonstrations included agricultural use, where a farmer leveraged ChatGPT to manage large-scale operations, improve efficiency, and coordinate team workflows. The example underscored the technology’s growing role in non-technical and field-based industries.
The release of GPT‑5.6 and its associated tools marks a shift toward AI systems that actively execute complex tasks, signaling a move from conversational assistants to integrated digital coworkers across industries.