OpenAI has announced a series of transformative updates today, headlined by the official release of the GPT-5.5 model series and the introduction of autonomous "Workspace Agents." These updates represent a fundamental shift in the company’s direction, moving beyond interactive chat interfaces toward systems capable of independent planning and cross-tool execution. Key updates include the launch of GPT-5.5 and GPT-5.5 Pro, the debut of ChatGPT for Clinicians, specialized tools for healthcare professionals, and a revamped subscription structure featuring a new $100-per-month Pro tier.

The GPT-5.5 Series Marks the Advent of Agentic Reasoning

The launch of GPT-5.5 and its high-performance counterpart, GPT-5.5 Pro, stands as the most significant milestone in OpenAI’s recent roadmap. Unlike previous generational leaps that focused primarily on context window size or conversational fluidity, GPT-5.5 is engineered specifically for "agentic" workflows. This means the model is optimized not just to answer questions, but to function as an autonomous agent that can navigate complex, multi-step tasks with minimal human intervention.

Intelligence Without Sacrificing Latency

One of the most notable technical achievements of GPT-5.5 is its efficiency. Typically, increasing a model's intelligence results in slower inference speeds. However, GPT-5.5 matches the per-token latency of the previous GPT-5.4 model while delivering a substantial step-up in reasoning capabilities. In real-world serving, it uses significantly fewer tokens to complete complex tasks in Codex, making it both more capable and more cost-effective for enterprise-scale deployments.

The model is designed to "plan, execute, and refine" independently. When given a goal—such as "set up a CI/CD pipeline for this repository and debug any environment errors"—GPT-5.5 does not simply provide instructions. Instead, it creates a mental map of the required steps, uses available tools to execute them, checks its own work against error logs, and iterates until the task is finished.

Benchmarking the New Frontier of Autonomy

The performance gains of GPT-5.5 are most visible in coding and professional reasoning benchmarks. On Terminal-Bench 2.0, a rigorous test for complex command-line workflows requiring planning and tool coordination, GPT-5.5 achieved a state-of-the-art accuracy of 82.7%.

In software engineering tasks, the model has set new records:

  • SWE-bench Pro: GPT-5.5 reached 58.6% in resolving real-world GitHub issues end-to-end.
  • Expert-SWE: On long-horizon coding tasks with a median human completion time of 20 hours, GPT-5.5 consistently outperformed GPT-5.4, demonstrating what early testers describe as "conceptual clarity."

Testers from organizations like NVIDIA and Magic Path noted that GPT-5.5 displays an intuitive understanding of system architecture. In one instance, the model successfully merged a branch with hundreds of frontend refactor changes into a main branch that had also undergone significant changes, resolving the merge in a single 20-minute pass.

Workspace Agents and the Automation of Knowledge Work

Accompanying the new models is the launch of "Workspace Agents," a cloud-based feature available to Business, Enterprise, Education, and Teacher subscribers. These agents represent the practical application of the GPT-5.5 agentic core, designed to run in the background to handle long-running workflows that previously required constant human oversight.

Autonomous Background Execution

Workspace Agents differ from standard GPT interactions in their persistence. A user can delegate a task—such as "monitor these five Slack channels for project updates and summarize them into a weekly report in Notion"—and the agent will execute the workflow autonomously in the cloud. These agents are built to:

  • Gather Context Across Systems: They can access and synthesize data from multiple connected apps like Slack, Gmail, Google Drive, and Microsoft Outlook.
  • Handle Ambiguity: If an agent encounters an obstacle, it is trained to reason through the failure and attempt an alternative path rather than stopping and asking for help.
  • Collaborate Within Teams: Organizations can now build specialized agents and share them within their private workspace, creating a library of digital teammates tailored to specific business functions.

Pricing and Availability of Agents

OpenAI has placed Workspace Agents in a research preview phase. Access is currently free for eligible paid tiers until May 6, 2026. Following this period, the feature will transition to a credit-based pricing model, allowing enterprises to scale their use of background agents based on the complexity and volume of the tasks being performed.

Specialized Tools for Healthcare Professionals

In a move toward industry-specific AI, OpenAI has launched "ChatGPT for Clinicians." This specialized workspace is available at no cost to verified U.S. healthcare professionals, including physicians, nurses, and pharmacists.

Bridging AI and Clinical Practice

ChatGPT for Clinicians is not a general-purpose chatbot but a clinical support tool integrated with research databases. It features:

  • Clinical Search: Enhanced capabilities to retrieve peer-reviewed medical literature and clinical guidelines.
  • Documentation Support: Tools designed to assist with medical record drafting and clinical note-taking, adhering to strict data privacy standards.
  • CME Integration: The platform allows professionals to earn Continuing Medical Education (CME) credits through their research and learning activities on the site.

To support the development of medical AI, OpenAI also released HealthBench Professional, an open benchmark designed to evaluate how large language models (LLMs) perform on clinical tasks. This benchmark aims to provide a standardized metric for safety and accuracy in the high-stakes environment of healthcare.

Expansion of Subscription Tiers and Codex Usage

To accommodate the increased compute demands of agentic workflows, OpenAI has updated its subscription options. The most notable change is the introduction of a new $100-per-month Pro plan.

The New Pro Tier vs. Plus

The $100/month Pro plan is specifically designed for high-intensity users, such as developers and data scientists who rely on extended Codex sessions.

  • Pro ($100/month): Provides unlimited access to GPT-5.4 and access to the new GPT-5.5 Pro model. It includes 10x the standard Codex usage allowance compared to the Plus plan, making it the preferred choice for those running complex, long-duration engineering tasks.
  • Pro ($200/month): This existing tier remains the highest-usage option, continuing its current Codex promotions through May 31, 2026.
  • Plus ($20/month): Remains the standard for day-to-day use. However, OpenAI has rebalanced Plus usage to support more sessions throughout the week rather than concentrated, high-intensity sessions in a single day.

Functional Enhancements for Daily Productivity

Beyond the major model releases, OpenAI has rolled out several updates aimed at making ChatGPT more integrated into the user's physical and digital environment.

Apple CarPlay Integration

OpenAI is rolling out ChatGPT for Apple CarPlay, providing a hands-free voice experience for users on the go. Compatible with iOS 26.4 or newer, this feature allows users to start new voice conversations or resume existing ones directly from the vehicle’s interface. This is a significant step for productivity, allowing users to brainstorm ideas, listen to summaries, or manage tasks while commuting.

Outlook Shared Mailboxes and Calendars

For professional users, ChatGPT now supports delegated and shared Outlook resources. If a user has the appropriate Microsoft permissions, they can ask ChatGPT to:

  • Read and browse messages in a shared mailbox.
  • Move messages between folders or mark them as read/unread.
  • Send plain-text emails from the shared address.
  • Manage shared calendars by creating, updating, or deleting events and handling RSVPs.

This integration requires connecting the Outlook Email or Outlook Calendar app within the ChatGPT interface and specifying the exact shared mailbox address during the prompt.

Simplified File Management and Large Pastes

Handling large amounts of data has become more streamlined with two key updates:

  1. Automatic Attachments for Large Pastes: When a user pastes more than 5,000 characters into the chat composer, ChatGPT now automatically converts that text into an attachment. This prevents the UI from becoming cluttered and preserves the context window’s efficiency.
  2. Google Drive App Unification: OpenAI has consolidated its Google file connectors into a single Google Drive app. This allows users to interact with Docs, Sheets, and Slides through a single authentication point, simplifying the workflow for those who rely on the Google Workspace ecosystem.

Advanced Creativity with Image 2.0

OpenAI has also updated its visual capabilities with the launch of Image 2.0. This new image generation model is integrated directly into ChatGPT and offers:

  • Superior Text Rendering: Previous models often struggled with legible text within images; Image 2.0 significantly improves this, allowing for the creation of posters, diagrams, and UI mockups with accurate spelling.
  • Multilingual Support: The model can now generate text in various languages and scripts with higher precision.
  • Advanced Visual Reasoning: The model can better interpret complex prompts that require an understanding of spatial relationships and artistic styles.

Safety Guardrails and Policy Updates

As AI models become more autonomous, the risks associated with their use evolve. OpenAI has introduced several new safety measures specifically for the GPT-5.5 era.

Bio Bug Bounty Program

To address risks related to biological threats, OpenAI launched the GPT-5.5 Bio Bug Bounty Program. This "red-teaming" challenge invites security researchers to find "jailbreak" methods that could bypass safeguards regarding biological risks. The program offers rewards of up to $25,000 for verified vulnerabilities, demonstrating a proactive approach to preventing the misuse of the model’s advanced scientific reasoning.

Cybersecurity and Government Briefings

OpenAI has been briefing government officials on the capabilities of GPT-5.5 concerning cybersecurity. The company is proposing a "dual-track" access model. Under this framework, public-facing versions of the model remain strictly safeguarded to prevent misuse in creating malware. Simultaneously, more permissive versions are provided to vetted security defenders, allowing them to use the model's advanced reasoning to find and patch vulnerabilities in critical infrastructure.

The OpenAI Privacy Filter

For developers and enterprises concerned with data privacy, OpenAI has released the OpenAI Privacy Filter. This is an open-weight model designed to detect and redact personally identifiable information (PII) from text with state-of-the-art accuracy. This tool is part of a broader effort to ensure that as AI integrates more deeply into business workflows, sensitive data remains protected.

Summary of Today's OpenAI Updates

The updates released today mark a transition from AI as a passive assistant to AI as an active participant in work and research. The launch of GPT-5.5 and GPT-5.5 Pro provides the "brain" for this new era, while Workspace Agents and specialized tools for clinicians provide the "hands" to execute tasks across different environments.

  • Models: GPT-5.5 and 5.5 Pro are now rolling out to Plus, Pro, Business, and Enterprise users.
  • Agents: Workspace Agents are in research preview for paid tiers, enabling autonomous background tasks.
  • Industry: ChatGPT for Clinicians offers a verified, no-cost workspace for medical professionals.
  • Integrations: Enhanced support for Apple CarPlay, Outlook shared mailboxes, and a unified Google Drive experience.
  • Safety: New Bio Bug Bounty program and the release of an open-weight Privacy Filter.

Frequently Asked Questions

What is the difference between GPT-5.5 and GPT-5.5 Pro?

GPT-5.5 is the standard high-intelligence model designed for agentic workflows, available to Plus and Pro users. GPT-5.5 Pro is an optimized version with higher reasoning limits and better performance on extremely complex coding and scientific tasks, primarily targeted at Pro, Business, and Enterprise subscribers.

How do Workspace Agents work?

Workspace Agents run in the cloud and can perform multi-step tasks independently. Unlike a standard chat, they can navigate across different apps (like Slack and Gmail) to complete a workflow without the user needing to provide a prompt for every step.

Who can access ChatGPT for Clinicians?

This feature is currently available to verified U.S. healthcare professionals, including physicians, pharmacists, and nurses. Verification is required to ensure the tools are used within a professional medical context.

What is the new $100/month Pro plan?

This plan is a middle ground between the $20 Plus plan and the $200 high-usage Pro plan. It is designed for professional developers and power users who need significant Codex usage and access to the latest GPT-5.5 Pro models but do not require the absolute highest usage limits of the $200 tier.

Is GPT-5.5 available via API?

Full API availability for GPT-5.5 and GPT-5.5 Pro is currently delayed as OpenAI works with partners on specific safety and security requirements for serving agentic models at scale. It is expected to be released to the API "very soon."

How does the Bio Bug Bounty program work?

Researchers can participate in a red-teaming challenge to find ways to bypass the model's safety filters regarding biological risks. Successful "jailbreaks" that demonstrate a legitimate risk can earn the researcher up to $25,000.