The latest AI news and updates of OpenAI — the AI research company behind ChatGPT, GPT models, Codex, GPT Image, and the OpenAI API. Covering OpenAI's latest product updates, company news, and launches from the past 90 days.
OpenAI15h agoOpenAI launched an AI agent on its developer documentation site to help builders find answers and navigate technical resources. The agent provides direct links to relevant documentation and generates custom guides with tailored prompts, which developers can open directly in Codex or copy as Markdown for use in other coding agents.
OpenAI15h agoOpenAI added a developer mode to the Codex app, enabling direct use of the Chrome DevTools Protocol for browser debugging. The feature allows Codex to profile JavaScript performance, inspect console logs, analyze network traffic, and examine live page state. Developers can enable full CDP access in browser settings to diagnose issues within the in-app browser or Chrome.
OpenAI15h agoOpenAI is rolling out the ability for Go, Plus, Pro, and Business plans to save Codex rate limit resets for later use, starting with one free reset. A two-week referral program allows Plus and Pro accounts to earn additional banked resets when invited friends send their first Codex message.
OpenAIJun 10OpenAI has updated its Responses API to support image results in addition to text when using the web search tool. This allows developers to build applications that can surface visual content like products, places, and visual references directly from the web. The update enhances the multimodal capabilities of AI applications by providing richer, visually-grounded responses.
ChatGPTJun 9OpenAI's ChatGPT now generates charts from uploaded data directly within the chat interface. This update brings data visualization capabilities to both mobile and web, enhancing in-chat analysis for users.
ChatGPTJun 7ChatGPT on the web now allows users to send emails directly from writing blocks within conversations. This streamlines workflows by integrating the final action of sending an email directly into the AI's content generation environment, reducing context switching.
ChatGPTJun 7OpenAI is rolling out an enhanced memory system for ChatGPT that automatically synthesizes and updates user preferences, constraints, and conversational context. This new architecture, built on "dreaming," allows ChatGPT to provide more personalized and continuous interactions, with users able to review and manage what it remembers.
OpenAIJun 5OpenAI has launched personal activity profiles within its Codex agentic coding platform. These profiles allow users to track their usage, including token consumption and feature adoption, and share highlights with others.
OpenAIJun 5OpenAI has released a Python SDK for its Codex agentic coding tool. This allows developers to programmatically integrate Codex's capabilities for complex engineering tasks into their own applications, custom agents, and CI/CD pipelines.
OpenAIJun 4OpenAI's Codex now features a "Build iOS Apps plugin" that enables viewing, testing, and hot reloading iOS app edits directly within Codex. This integration allows autonomous agents to interact with iOS apps in an in-app browser and SwiftUI previews.
OpenAIJun 4OpenAI now provides moderation scores directly within its Responses API and Completions API. This allows developers to get safety signals for both input and generated content in a single request, simplifying the integration of content policies into AI applications.
OpenAIJun 4OpenAI is rolling out a more capable memory system for ChatGPT that automatically curates and updates context across conversations. This new system, built on "dreaming" architecture, helps ChatGPT remember user preferences and adapt to the passage of time, making interactions more personalized and relevant.
OpenAIJun 4OpenAI updated its GPT-Rosalind model series to include agentic coding and tool-use features from GPT-5.5. The update adds specialized intelligence for medicinal chemistry and genomics to automate complex experimental workflows. This shift allows researchers to move from simple data analysis to autonomous hypothesis testing and lab troubleshooting.
OpenAIJun 2OpenAI has launched specialized Codex plugins that bundle 62 applications and 110 skills for non-technical roles like sales and finance. The update introduces interactive workspaces called Sites and inline refinement through Annotations to help teams collaborate without writing code.
OpenAIJun 2OpenAI has launched Sites for Codex, allowing users to generate and share interactive web applications from text prompts. This update shifts Codex from a specialized coding tool into a no-code platform for non-technical professional workflows.
OpenAIJun 2OpenAI has moved its frontier models and the Codex coding agent into general availability on Amazon Bedrock. This allows organizations to run OpenAI's most capable reasoning and automation tools within their existing AWS security and compliance perimeters.
OpenAIMay 30Tibo, the Codex engineering lead at OpenAI, clarified that each decimal increment in the GPT-5 series represents simultaneous gains in model capability and token efficiency. This strategy ensures that newer versions like GPT-5.5 deliver higher intelligence without the speed penalties typically associated with larger models.
OpenAIMay 30OpenAI introduced visual identicons and a global search feature to the Codex app to help developers manage complex multi-agent workflows. These updates provide visual consistency for background agents and make it easier to retrieve context from past development threads.
OpenAIMay 29OpenAI expanded its Computer Use capability to the Codex desktop app for Windows, allowing AI agents to operate graphical interfaces, change settings, and test software natively. Users can now steer these autonomous desktop tasks remotely via the ChatGPT mobile app while work continues on their Windows machine.
OpenAIMay 29OpenAI launched Rosalind Biodefense to provide sponsored access and support for vetted developers and government agencies building pandemic preparedness tools. This initiative shifts frontier AI from a general research tool to a restricted defensive asset, mirroring OpenAI's recent strategy of arming defenders in critical infrastructure sectors.
OpenAIMay 27OpenAI released Secure MCP Tunnel, a networking tool that lets AI agents access private on-premises databases and tools through outbound-only HTTPS connections. This removes the need for inbound firewall exceptions, allowing enterprises to safely ground agents in sensitive internal data.
OpenAIMay 27OpenAI will retire GPT-5.2 and GPT-5.3-Codex from the Codex web interface on June 2nd, making GPT-5.5 the new default for free users. This shift reflects the rapid obsolescence of frontier models, as the five-month-old GPT-5.2 now accounts for less than 1% of production traffic.
OpenAIMay 25OpenAI's Codex agent can now autonomously manage personal administrative toil like identifying and canceling unwanted paid subscriptions. By scanning emails and using its computer control capabilities to navigate cancellation flows, the agent moves beyond software development into general-purpose professional automation.
OpenAIMay 23OpenAI resolved a technical issue in Codex that caused user usage limits to deplete faster than intended during long-running sessions. The team rolled back a faulty compaction optimization and performed a full reset of usage limits for all accounts to restore developer access.
OpenAIMay 22OpenAI updated the Codex desktop app to allow its Computer Use feature to operate while a Mac is locked and the screen is off. By installing a specialized authorization plug-in, the agent can now perform long-running background tasks triggered remotely from a mobile device.
OpenAIMay 21OpenAI rolled out a batch of updates to its Codex agentic coding platform. Goal Mode runs toward objectives across hours or days, Appshots binds Command-Command to capture full-window context on Mac, and advanced browser annotation lets designers direct UI changes inline.
OpenAIMay 20An internal OpenAI reasoning model autonomously solved the planar unit distance problem, disproving a 1946 conjecture by Paul Erdős. Researchers Alex Wei, Hongxun Wu, and @wjmzbmr1 detailed the breakthrough and the future of human-AI collaboration on the OpenAI Podcast.
OpenAIMay 19OpenAI introduced Guaranteed Capacity, a reservation program that allows enterprise customers to secure long-term access to compute through one-to-three-year commitments. This shift from on-demand usage provides the infrastructure certainty required to scale production-grade AI agents and critical customer workflows.
OpenAIMay 19OpenAI is implementing a multi-layered provenance system that combines C2PA metadata with Google's SynthID invisible watermarks to identify AI-generated images. This shift moves beyond fragile metadata to a more resilient verification model that survives file edits and platform transfers.
OpenAIMay 18OpenAI released Remote Connections for the Codex desktop app, allowing users to monitor and control autonomous coding sessions via the ChatGPT mobile app. The update enables developers to approve terminal commands and review code diffs from their phones while the agent runs on a host Mac or SSH environment.
OpenAIMay 16OpenAI updated its Codex agentic coding platform with fully customizable keyboard shortcuts and a revamped local server management interface. The release introduces significant performance gains for large repositories, reducing re-rendering by 75% and accelerating expensive Git operations by up to 50 times.
ChatGPTMay 16OpenAI launched a preview for ChatGPT Pro users in the U.S. that allows them to securely link financial accounts to the assistant. By grounding conversations in real-time transaction data and portfolio performance, the system provides personalized budgeting and long-term financial modeling.
OpenAIMay 16OpenAI resolved two technical issues that caused a 48-hour degradation in GPT-5.5's reasoning and execution capabilities within the Codex platform. To compensate for the downtime, the company is performing a manual reset of usage limits for all affected users this evening.
OpenAIMay 15OpenAI is investigating user reports that GPT-5.5 is performing worse than expected within the Codex coding platform. While internal systems are currently healthy, the engineering team is looking for the cause of the perceived performance dip in its newest frontier model.
OpenAIMay 15OpenAI released a preview of Codex in the ChatGPT mobile app, allowing users to monitor and approve autonomous coding tasks from iOS and Android devices. This update shifts agentic workflows from desk-bound sessions to persistent, mobile-steerable loops that keep work moving across remote environments.
OpenAIMay 14OpenAI developed a custom security sandbox for Codex on Windows to enable autonomous command execution without compromising system safety. The new architecture uses dedicated local users and firewall rules to strictly control file access and network traffic for coding agents.
OpenAIMay 13OpenAI launched a 30-day promotion offering two free months of Codex usage for new users at eligible enterprise customers that switch their teams to the platform during the window. The pitch in OpenAI's announcement is aimed at developers: send the post to your CTO to formalize Codex use at work.
OpenAIMay 13OpenAI updated Computer Use in the Codex desktop app so agents can click, type, and execute tasks across applications without taking over the user's Mac. This enables AI to work alongside humans on the same screen and cursor. Developers now handle complex multi-app workflows without workflow interruptions.
OpenAIMay 12OpenAI launched Daybreak, a cybersecurity platform that integrates GPT-5.5 models and Codex to identify, validate, and remediate software vulnerabilities. The system shifts security teams from manual triage to autonomous agentic workflows that can generate and test code patches directly in repositories.
OpenAIMay 11OpenAI launched the OpenAI Deployment Company, a $4 billion standalone entity designed to help organizations move frontier AI into production. By acquiring engineering firm Tomoro and partnering with 19 global investment and consulting firms, the company will embed specialized engineers to redesign business workflows around reasoning models.
OpenAIMay 8OpenAI released a guided migration tool for its Codex app that automatically imports configurations and skills from other AI coding agents. The update targets developers facing rate limits or setup friction by turning personalized agent environments into portable assets.
OpenAIMay 8OpenAI discovered that several released models were accidentally rewarded for their internal reasoning steps during training, a practice usually avoided to prevent AI from learning to hide its thoughts. Analysis of the affected runs showed no measurable drop in the models' honesty, though the company is implementing new automated safeguards to prevent future leaks.
OpenAIMay 8OpenAI published a technical guide for gpt-realtime-2 that introduces granular controls for reasoning effort and spoken preambles. This shift allows developers to tune the balance between voice latency and complex problem-solving for autonomous audio interactions.
OpenAIMay 7OpenAI released GPT-Realtime-2 alongside new streaming translation and transcription models in its Realtime API. This update shifts voice AI from simple conversational loops to reasoning-capable agents that can solve complex problems and handle interruptions in real time.
OpenAIMay 7OpenAI released a Codex Chrome extension for macOS and Windows that executes multi-step workflows across browser tabs in the background. By using the user's authenticated browser state, the agent can now manage tasks on private dashboards and internal tools without interrupting active work.
OpenAIMay 7OpenAI released Multipath Reliable Connection (MRC) as an open networking protocol to prevent single link failures from crashing massive AI training jobs. By spraying data across hundreds of paths and using static source routing, the protocol ensures frontier model training remains efficient even as clusters scale past 100,000 GPUs.
OpenAIMay 5OpenAI is rolling out GPT-5.5 Instant as the default model for all ChatGPT users, featuring a more concise and natural tone. The update integrates context from connected Gmail accounts and saved memories while adding a transparency layer to show which data sources influenced a response.
OpenAIMay 5OpenAI released the /goal command for its Codex agentic coding platform, enabling users to define high-level objectives that the AI pursues autonomously. This update shifts the developer's role from issuing iterative prompts to providing strategic instructions for long-horizon tasks that can span multiple files and days.
OpenAIMay 5OpenAI rearchitected its media infrastructure with a split relay and transceiver model to support 900 million weekly users on Kubernetes. By routing packets based on protocol metadata rather than dedicated ports, the system maintains sub-second latency for real-time voice interactions at global scale.
OpenAIMay 1OpenAI's API revenue is growing twice as fast as any previous release just one week after the debut of GPT-5.5. This surge, alongside a doubling of Codex revenue, signals a massive enterprise shift toward autonomous agentic coding workflows.
OpenAIMay 1OpenAI introduced Codex Pets, animated virtual companions that provide glanceable status updates for background agent threads within the Codex desktop app. This shift adds a visual monitoring layer to autonomous coding loops, allowing developers to track progress without switching windows.
OpenAIMay 1OpenAI released a migration feature for the Codex app and CLI that imports configurations, plugins, and agent skills from other AI coding tools. By lowering the friction of switching platforms, OpenAI is targeting developers who have invested time in customizing rival agents. This move shifts the competition from raw model performance to the portability of a developer's personalized agent environment.
OpenAIApr 30OpenAI released Advanced Account Security, an opt-in setting that mandates phishing-resistant sign-in and hardened account recovery for ChatGPT and Codex. This shift moves AI accounts from simple chat profiles to critical infrastructure that requires hardware-backed protection against sophisticated digital attacks.
OpenAIApr 30OpenAI released Codex for Work, a personal assistant that connects to business tools like Slack and Google Drive to automate recurring tasks and research. This shift moves Codex from a specialized coding tool to a general-purpose agent that manages professional toil across a team's entire tech stack.
OpenAIApr 30OpenAI published a technical post-mortem tracing the goblin behavioral quirk in GPT-5 models to unintended reinforcement during personality training. The investigation reveals how a specific reward signal for a playful persona leaked into the base model behavior, creating a persistent feedback loop.
OpenAIApr 30OpenAI says its Codex agentic coding platform can already work for days on hard tasks with small configuration tweaks. The team plans to release changes that make these long-running workflows easier for everyone, signaling a near-term move toward multi-day autonomous engineering sessions.
OpenAIApr 30OpenAI introduced a limited-time $0 seat fee for Codex-only users on ChatGPT Business and Enterprise plans through June. This shift to usage-based billing allows teams to pilot autonomous coding agents without fixed monthly costs while scaling spend directly with token consumption.
OpenAIApr 30OpenAI introduced WebSocket support for its Responses API to eliminate the latency overhead of traditional HTTP requests in multi-step agent workflows. By maintaining a persistent connection and caching conversation state in memory, the system allows coding agents to run up to 40% faster end to end.
OpenAIApr 30OpenAI expanded its Codex agentic coding platform to autonomously generate architecture diagrams and implementation plans within FigJam. This shift moves AI assistance from pure code generation to the visual planning phase of the software development lifecycle.
OpenAIApr 29OpenAI updated its Codex terminal agent to allow developers to migrate entire codebases to the GPT-5.5 model through a single prompt. This shift automates the refactoring and dependency updates required to adopt new frontier models, reducing the manual effort of staying current with API changes.
Frequently asked questions