OpenAI just launched the AI coworker era

What happened

Earlier this week, OpenAI released GPT 5.6, which includes a new flagship model named Sol. This release introduces three models: Luna, described as an affordable, everyday version; Terra, positioned as a middle tier balancing power and cost; and Sol, the top-of-the-line model built for sustained, complex work.

This update represents a shift in AI's function. Previously, AI models primarily acted as answering machines, waiting for explicit instructions for each step. GPT 5.6 is designed for AI to work on behalf of users for extended periods across multiple tools without constant hand-holding.

Sol achieved a score of 53.6 on the Agent's Last Exam, a benchmark assessing AI performance on long professional workflows across 55 fields. This score was over 13 points higher than the closest competitor. Additionally, Sol scored 62.6% on OS World 2.0, a test measuring an AI's ability to control a computer, while using 85% fewer words than a competing model.

Why it matters

This release signals a fundamental redefinition of AI's role in the workplace. AI is moving beyond simple query-response interactions to function as a collaborative junior colleague. Models like Sol are designed to take on multi-step projects, browse the web, utilize tools, and even write code, delivering finished work. This capability automates complex workflows that previously required significant human effort.

For businesses and individuals, this means a potential for substantial productivity gains. Affordable models like Luna can manage routine tasks such as emails or social media posts, freeing up hours of work. Terra can support research and organization for more complex material. The flagship Sol can perform competitive analyses or other extensive projects, allowing managers to delegate entire tasks rather than just components.

However, this shift also carries implications for the human workforce. As AI becomes capable of managing professional workflows that historically required analysts or junior associates, companies will evaluate the economic impact. This environment will prioritize individuals who learn to effectively work with and supervise these AI tools, rather than those who ignore their capabilities.

What to watch next

• How businesses begin integrating these new agentic AI models into their core workflows.
• The speed at which other AI developers release competitive models with similar multi-step, autonomous capabilities.
• Changes in job descriptions and required skill sets as organizations adapt to AI agents taking on more complex tasks.
• Whether the performance gaps highlighted by OpenAI's benchmarks persist or narrow as the technology evolves.
• The availability and pricing of these advanced models across various platforms and API tiers.

What this means for you

For business leaders and operators, the primary takeaway is to adjust your approach to AI utilization. Stop treating AI as merely a search engine or a single-query tool. Instead, begin delegating multi-step projects to AI models. Identify tasks that involve several steps and typically consume hours of manual effort, then provide the AI with the project scope, known information, and constraints, and allow it to execute.

This transition requires developing a new skill: supervising AI. You will need to provide clear direction, check the AI's work for accuracy, and iterate on its outputs. This practice moves beyond simple chatbot interactions to effectively manage an AI agent. Acquiring proficiency in this area will provide a distinct advantage as the workplace evolves to incorporate increasingly capable AI coworkers.

Key takeaways

• OpenAI released GPT 5.6 with new models Luna, Terra, and the flagship Sol.
• This update shifts AI from a question-answering tool to a multi-step project collaborator.
• Sol demonstrated superior performance on benchmarks for complex professional workflows.
• Learning to supervise AI agents is becoming a critical professional skill.
• Delegating multi-step projects to AI, rather than just asking questions, is the new paradigm.

What is OpenAI's GPT 5.6?

OpenAI's GPT 5.6 is a new family of AI models released by OpenAI, featuring Luna, Terra, and the flagship Sol. This release marks a significant shift in AI capability, moving beyond simple question answering to enable AI to work on behalf of users for extended periods across multiple tools. The models are designed for various use cases, from affordable everyday tasks to complex, sustained professional work.

What are the new AI models introduced with GPT 5.6?

With GPT 5.6, OpenAI introduced three new models: Luna, Terra, and Sol. Luna is described as the affordable, everyday version, suitable for cost-sensitive tasks. Terra is the middle-tier option, offering a balance between power and cost. Sol is the flagship model, designed for serious, sustained, and complex professional workflows. These models aim to cater to different levels of need and computational intensity.

How does Sol perform on AI benchmarks?

Sol, the flagship model in OpenAI's GPT 5.6 family, demonstrated strong performance on key benchmarks. On the Agent's Last Exam, which tests AI's ability to handle long professional workflows across 55 different fields, Sol scored 53.6. This was more than 13 points higher than its closest competitor. Additionally, Sol scored 62.6% on OS World 2.0, a test measuring an AI's capacity to control a computer, while using 85% fewer words than a competing model.

How is AI's role changing with these new models?

The role of AI is changing from a simple answering machine or search engine to that of a capable junior colleague or collaborator. With models like Sol, AI is designed to take on multi-step projects, browse the web for information, use various tools, and even write code, all without needing constant instruction at every step. This enables AI to execute complex tasks and deliver finished work, allowing users to delegate projects rather than just ask questions.

What skill is becoming important due to advanced AI?

The skill of supervising AI is becoming increasingly important. As AI models become capable of handling complex, multi-step professional workflows, individuals need to learn how to give clear directions to AI, check its work for accuracy, and iterate on its outputs. This ability to effectively manage and guide AI agents, rather than just using chatbots for simple queries, is becoming a critical professional skill, similar to communication or project management.