OpenAI Dev Day 2025: Apps SDK, AgentKit, and New Models Transform Developer Platform
OpenAI held its third annual Dev Day conference at Fort Mason in San Francisco, unveiling a suite of new tools and models aimed at solidifying its position as a comprehensive AI development platform.
Platform Growth
CEO Sam Altman opened the event by highlighting OpenAI's explosive growth. The platform now serves 800 million weekly ChatGPT users, up from 100 million in 2023. The developer community has doubled to 4 million, while API usage has surged from 300 million tokens per minute to 6 billion.
Three Major Announcements
1. Apps in ChatGPT with Apps SDK
OpenAI introduced the Apps SDK, allowing developers to build interactive applications that run directly inside ChatGPT conversations. Built on the Model Context Protocol (MCP), the SDK enables apps to trigger actions and render full user interfaces within the chatbot.
The company demonstrated live integrations with Canva, Spotify, and Coursera during the keynote. Developers can now reach hundreds of millions of ChatGPT users, with planned monetization features including a new e-commerce checkout protocol.
2. AgentKit: Complete Agent-Building Platform
AgentKit provides developers with an end-to-end toolkit for building, deploying, and optimizing AI agents. The platform includes:
- Agent Builder: A visual, drag-and-drop workflow creator
- ChatKit: Embeddable chat interfaces for custom applications
- Guardrails: Safety screening for inputs and outputs
- Evals: Datasets, trace grading, and automated prompt optimization
- Connectors: Integration with various data sources
The no-code approach aims to make agent development accessible to developers without requiring extensive AI expertise.
3. New Models in the API
OpenAI released several new models for developers:
- GPT-5 Pro: Described as the "smartest model in the API," designed for tasks requiring high accuracy and deep reasoning. Pricing is set at $15 per million input tokens and $120 per million output tokens.
- Sora 2: The company's next-generation video and audio generation model, now available via API. Sora 2 creates videos with synchronized dialogue, realistic soundscapes, and improved physics simulation. The model allows detailed control over video length, aspect ratio, and resolution.
- gpt-realtime-mini: A smaller, more affordable voice model offering 70% cost savings compared to the full voice model while maintaining quality. Altman emphasized that voice will become "one of the primary ways people interact with AI."
Additional Updates
OpenAI announced that Codex, its AI-powered coding assistant, has moved from preview to general availability. The tool now includes enterprise controls, Slack integration, and a Codex SDK. Usage has increased tenfold since its August launch, with over 40 trillion tokens served.
The company also revealed a multi-year partnership with AMD to deploy 6 gigawatts of Instinct GPUs, addressing the ongoing compute capacity challenges facing AI development.
Looking Forward
In a closing conversation with former Apple design chief Jony Ive, who joined OpenAI through its acquisition of his startup io earlier this year, the pair discussed future hardware ambitions. Ive mentioned the team is exploring 15-20 AI device concepts focused on joy, connection, and emotional balance rather than pure productivity.
When asked about profitability concerns, Altman told reporters that revenue generation is "not in his top 10 concerns," emphasizing that OpenAI is focused on investment and growth for the foreseeable future.
The announcements position OpenAI as not just an AI model provider, but as a comprehensive platform for developers to build, deploy, and monetize AI-powered applications at scale.
