OpenAI has fundamentally restructured ChatGPT with three distinct work modes, new models including GPT-5.6, and a rebuilt desktop application that mirrors features users already know from competitors like Claude. The overhaul includes Chat for quick conversations, Work for longer-form deliverables, and Codex for direct integration with local folders and files. The changes rolled out gradually across web, mobile, and desktop platforms, with availability depending on user plan, region, and device type.

What Are the Three New ChatGPT Work Modes?

OpenAI introduced three separate interfaces within the rebuilt ChatGPT ecosystem, each designed for different workflows. Chat remains the familiar conversational interface for quick questions and back-and-forth dialogue. Work is a new agent that handles longer tasks and produces finished deliverables like documents, spreadsheets, presentations, and reports. Codex operates directly with local folders, repositories, and repeatable systems, allowing users to work with files and apps on their computers.

The desktop app now brings all three modes into one unified experience on macOS and Windows, though some features roll out at different times depending on your plan and region. The web version at chatgpt.com remains closer to the original ChatGPT interface but now includes a Work tab so users can switch between Chat and Work modes without leaving the browser. Mobile serves as a control surface for voice, camera, and quick questions, with the ability to check on work running elsewhere, such as Codex processes on a desktop computer.

How Do the New GPT-5.6 Models Compare?

OpenAI introduced a new model family called GPT-5.6 across ChatGPT, Codex, and the API. The model lineup includes three tiers designed for different use cases. Sol is positioned as the smart, reasoning-focused model best for messy, open-ended tasks or work that matters significantly. Terra serves as the everyday model, replacing the previous GPT-5.5 for general-purpose work. Luna is the fast, cost-effective option designed for repetitive jobs where the correct answer is already known, such as data extraction or categorization.

Beyond model selection, users can now control effort levels, which determine how much computational thinking the model applies before answering. The effort options range from Light to Ultra, with Max giving the model extended time on a single task and Ultra splitting tasks into pieces and running them simultaneously. OpenAI's guidance suggests using the lowest effort level that achieves the desired result, avoiding unnecessary computational overhead.

What New Features Ship With the Rebuilt Desktop App?

• Projects: Organize conversations and work into dedicated project spaces for better workflow management
• Skills: Reusable workflows that teach ChatGPT to run a process the same way every time, enabling consistency across repeated tasks
• Templates: Pre-built starting points for common work types, reducing setup time for standard deliverables
• Plugins: Integrated apps and tools accessible through a redesigned Plugin Directory that packages apps, skills, and instructions into reusable workflows
• Sites: Ability to ask Work or Codex to build and publish working websites directly from the interface
• Scheduled Tasks: Automation features that run prompts on a schedule or monitor for changes without manual intervention
• Visualize: Tools for creating charts, graphs, and visual representations of data and analysis
• Pets: Customizable interface elements for personalization

The desktop app also includes GPT-Live, described as a significant upgrade to voice capabilities. The new models listen and talk simultaneously, allowing users to interrupt, pause to think, or wait until they finish speaking before responding. This represents a departure from previous turn-based voice interactions.

How Does ChatGPT Search Integrate Into the New System?

ChatGPT Search now pulls current information from the web with links to sources, ensuring answers are not limited to the model's training data cutoff. This feature addresses a long-standing limitation of large language models, which can only reference information they were trained on. By integrating live web search, ChatGPT can provide up-to-date responses on current events, recent research, and breaking news.

How to Set Up ChatGPT for Your Workflow

• Choose Your Model: Start with Sol for complex reasoning tasks, Terra for everyday work, or Luna for repetitive, straightforward jobs where the correct answer is already defined
• Set Effort Level: Begin with Medium effort for most tasks, then increase to High or Extra High only when the task requires complex reasoning or produces unsatisfactory results at lower levels
• Configure Memory and Data Controls: Access privacy settings and memory features on the web at chatgpt.com to control how ChatGPT remembers information across sessions and manages your data
• Create Skills for Repetition: If you perform the same task regularly, build a reusable Skill that teaches ChatGPT your preferred process, ensuring consistency without rewriting instructions each time
• Set Up Scheduled Tasks: For routine work like daily reports or monitoring tasks, configure Scheduled Tasks to run automatically on your preferred schedule

One important caveat: OpenAI is rolling out several features gradually. The buttons and options visible depend on your subscription plan, country, workspace settings, and whether you access ChatGPT via web, mobile, or desktop. Free and Go tier users have more limited access than paid subscribers.

What's the Practical Difference From Claude?

The rebuilt ChatGPT mirrors much of what already exists in the Claude ecosystem, close enough to feel like replicas of familiar workflows. However, early testing suggests GPT-5.6 Sol performs notably well on writing tasks, with some users reporting that it matched their voice and style better than Claude when given the same voice DNA skill instructions. The key difference lies in execution quality rather than feature parity; both platforms now offer similar architectural approaches to work modes, file integration, and task automation.

The speed of GPT-5.6 Sol on complex tasks like spreadsheet work and article polishing can be slower than some alternatives, with users reporting that medium-effort tasks sometimes take extended periods to complete. This suggests that reasoning depth comes at a computational cost, making Sol best suited for work where quality matters more than immediate turnaround.

What Does This Mean for IP and Legal Professionals?

The expansion of ChatGPT's capabilities arrives amid growing scrutiny of AI-generated content in professional settings. US judges have begun losing patience with lawyers who submit AI-generated fake citations in trademark and legal filings, signaling that AI tools require careful human oversight and verification. The new Work and Codex modes in ChatGPT could streamline legitimate legal research and document preparation, but professionals must remain vigilant about verifying all AI-generated content, particularly citations and factual claims that could face judicial scrutiny.