OpenAI Launches ChatGPT Work and GPT-5.6 for Enterprise

OpenAI has shifted its enterprise strategy from chasing raw benchmark leadership to optimizing performance per dollar. The company launched ChatGPT Work, an agentic platform designed to automate complex workplace tasks, alongside the general availability of its GPT-5.6 model family.

Unlike standard chat interfaces, ChatGPT Work operates across files and applications to execute long-running workflows. It can coordinate multiple tools to generate business documents, spreadsheets, and websites, moving the user experience from individual prompting to delegation.

This rollout follows a period of limited preview caused by US government restrictions. Regulators raised concerns regarding the models' advanced capabilities in biology and cybersecurity before the broader release was approved. According to computerworld.com, GPT-5.6 is now available via ChatGPT, Codex, and the OpenAI API.

Efficiency and Cost

The GPT-5.6 architecture focuses on maximizing the utility of every token to lower total operating costs. OpenAI is pitching this as a necessity for enterprises deploying artificial intelligence at scale, where inference expenses often outweigh the marginal gains of slightly higher accuracy.

The model lineup includes Sol, priced at $5 per million input tokens and $30 per million output tokens. For organizations requiring even lower costs for massive deployments, OpenAI introduced the Terra and Luna variants.

These models show improved performance in coding, scientific research, and enterprise knowledge work. By reducing token consumption while maintaining output quality, OpenAI aims to make agentic workflows financially viable for high-volume corporate use.

Market Context

This move comes as the cadence of frontier model releases has quadrupled since 2023, as tracked by aireleasetracker.com. The industry is moving away from the era of monolithic model updates toward a fragmented ecosystem of specialized, cost-optimized versions.

While OpenAI pushes for enterprise efficiency, competitors are focusing on different frontiers. Anthropic recently detailed a technique called the Jacobian lens to probe the internal states of its models, as reported by technologyreview.com. This research into mechanistic interpretability suggests a growing industry interest in not just what models can do, but how they arrive at their answers.

However, the gap between marketing and reliability remains. Even as OpenAI rolls out sophisticated enterprise agents, its latest voice model, GPT-Live-1, continues to struggle with basic logic tests. Recent reports from gizmodo.com highlight the model's failure to perform simple character counting, illustrating that reasoning gaps persist despite architectural leaps.

For the practitioner, the arrival of GPT-5.6 signals that the frontier is no longer just about the largest parameter count. The real competition has moved to the efficiency frontier, where the winner is the one who can deliver the most reliable agentic behavior for the lowest cost per request.

Can a focus on cost-efficiency sustain the lead in raw capability, or will the pursuit of the dollar compromise the intelligence of the next generation of agents?

FAQ

What is ChatGPT Work?
It is an agentic platform that uses GPT-5.6 to automate multi-step business tasks across different applications and files.

How much does GPT-5.6 Sol cost?
Sol is priced at $5 per million input tokens and $30 per million output tokens.

What are the Terra and Luna models?
These are lower-cost versions of the GPT-5.6 family designed for organizations scaling AI deployments.

Why was the rollout of GPT-5.6 delayed?
US government restrictions were imposed due to concerns over the model's capabilities in cybersecurity and biology.