Articles from openai.com

The piece describes practical strategies for using OpenAI's Codex as a persistent workspace to manage complex, long-running workflows across multiple prompts.

OpenAI announces a third-phase strategy to make AGI widely accessible, accelerate scientific research, and distribute economic gains broadly rather than concentrating power.

OpenAI launched a new memory system called Dreaming that automatically synthesizes context from past conversations to improve ChatGPT's personalization and freshness.

OpenAI launched role-specific plugins, Sites, and annotations for Codex, enabling non-developers to build apps, dashboards, and reports across 62 apps.

OpenAI and Thrive Holdings built a self-improving tax agent using Codex that automates tax preparation and measurably improves over time through practitioner feedback and production traces.

Codex is now available in the ChatGPT mobile app, enabling users to monitor and direct long-running coding tasks from their phone via a secure relay.

OpenAI launched GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper in its API to enable developers to build more natural, intelligent, and real-time voice applications.

OpenAI's GPT-5.5 Instant reduces hallucinated claims by 52.5% on high-stakes prompts and delivers tighter, more personalized responses as ChatGPT's new default model.

OpenAI rebuilt its WebRTC stack with a relay-transceiver architecture to reduce latency and manage scale for real-time voice AI across 900 million users.

OpenAI traces how their models' goblin metaphor tic originated from a reinforcement-learning reward signal for a Nerdy personality, then spread via training data contamination.

OpenAI launches workspace agents for team use, enabling shared, cloud-based agents that handle complex workflows like report generation and code writing within organizational permissions and controls.

OpenAI released a major update to Codex, enabling it to control the user's computer, use a browser, generate images, and automate tasks across the software development lifecycle.

OpenAI details its Model Spec as a public, evolving framework to define, train, and evaluate intended model behavior, balancing safety, user freedom, and accountability.

OpenAI's monitoring of internal coding agents using GPT-5.4 Thinking detects over-eager behavior but no scheming, with less than 0.1% of traffic outside coverage.

Codex Security was designed to start from repository context and intent rather than SAST reports, using validation and falsification to surface vulnerabilities with stronger evidence.

OpenAI argues that defending against prompt injection in AI agents requires treating attacks as social engineering and designing systems to constrain impact rather than just filtering inputs.

Companies that successfully reinvent with AI treat it as a portfolio of five compounding value models rather than isolated pilots, sequenced from workforce empowerment to agent-led process re-engineering.

OpenAI releases GPT-5.3 Instant, an update to ChatGPT's most-used model improving conversational tone, reducing refusals, and providing more accurate, well-synthesized answers.

OpenAI announces a contract with the Pentagon for classified AI deployments, emphasizing cloud-only deployment, retained safety guardrails, and explicit prohibitions against domestic surveillance and autonomous weapons.

OpenAI raises $110B at $730B valuation from SoftBank, NVIDIA, and Amazon to scale AI infrastructure and products.

OpenAI releases GPT-5.3-Codex-Spark, a real-time coding model with 1000+ tokens per second via Cerebras hardware, available in research preview for ChatGPT Pro users.

OpenAI's team built a production-grade codebase entirely via Codex agents, finding that engineering shifts from writing code to designing agent-legible environments, feedback loops, and architectural guardrails.

OpenAI launches GPT-5.3-Codex, a coding agent that improves on prior benchmarks, runs 25% faster, and was used to develop itself.

OpenAI introduced the Codex App Server, a bidirectional JSON-RPC protocol that exposes the full Codex agent harness for embedding in IDEs, web apps, and terminal clients.

OpenAI's new macOS Codex app serves as a command center for managing multiple AI coding agents in parallel, with skills, automations, and configurable sandbox security.

OpenAI built an internal AI data agent using GPT-5, Codex, and memory to reason over massive datasets, delivering insights in minutes.

OpenAI prevents URL-based data exfiltration by only allowing AI agents to auto-fetch URLs that were independently observed in a public web index.

OpenAI explains the architecture of Codex CLI's agent loop, detailing how it orchestrates model inference, tool calls, and context management using the Responses API.

OpenAI's CFO describes a business model where revenue scales with compute capacity, citing 10x ARR growth to $20B+ from 2023–2025.

OpenAI uses reinforcement learning-based automated red teaming to discover and patch prompt injection attacks against the ChatGPT Atlas browser agent.

OpenAI releases GPT-5.2-Codex, an agentic coding model with improved long-context understanding, Windows support, and cybersecurity capabilities, including discovery of a React vulnerability.

OpenAI releases GPT-Image-1.5, a new image generation model for ChatGPT and the API, claiming 4× faster generation with better editing and consistency.

OpenAI used its Codex agent to build the Sora Android app in 28 days with a four-engineer team, achieving a 99.9% crash-free rate on launch.

OpenAI CEO reflects on ten years of progress toward AGI, citing key breakthroughs and iterative deployment as the right strategy.

OpenAI launches shopping research in ChatGPT, a conversational tool that researches products and builds personalized buyer's guides using GPT-5 mini.

OpenAI advocates that "evals" — structured evaluation frameworks — are essential for businesses to turn vague AI goals into measurable, reliable outcomes.

OpenAI releases GPT-5.1-Codex-Max, a faster coding model using compaction for long-horizon tasks across multiple context windows.