From ChatGPT 5 Pro to Sora 2: Checkout the Biggest Announcements from OpenAI’s DevDay 2025

At OpenAI DevDay 2025, the company didn't just roll out new models, it laid the coordinates for its future platform ambitions. The announcements covered everything from next-level models and multimedia tools to embeddable developer SDKs, partnerships, and hints of future hardware.

In conversation with former Apple design chief Jony Ive, OpenAI also teased its vision of AI devices that change how we relate to technology.

Model Launches & Upgrades

GPT-5 Pro & Real-Time Voice

CEO Sam Altman unveiled GPT-5 Pro, positioning it as OpenAI's most precise and reasoning-focused language model to date. He also introduced gpt-realtime-mini, a cheaper low-latency voice model that enables real-time conversational experiences via voice. Both are now available to developers through the API.

Sora 2 & Multimedia AI

Another centerpiece was Sora 2, OpenAI's next-gen video + audio generation model. The upgrade enables richer soundscapes, synchronized audio, and more photorealistic visuals. Developers can access it via API, unlocking immersive video creation inside apps.

In tandem with the model, OpenAI also launched or updated a Sora app, where users can generate short vertical videos (à la TikTok style) using prompts.

Compact Image Model

To extend flexibility, OpenAI also delivered gpt-image-1-mini, a lighter, lower-cost image generation model to complement Sora 2.

Developer Tools & Platform Extensions

AgentKit, Apps SDK & ChatKit

OpenAI introduced AgentKit, a full toolkit for building, deploying, and optimizing AI agents, from prototype to production.

They also launched an Apps SDK (in preview), enabling third-party apps to embed and interact natively with ChatGPT. This turns ChatGPT into a kind of app platform, where you can "chat with apps" such as Spotify, Zillow, Canva, and more.

Complementing this is ChatKit (or embedding chat interfaces), allowing developers to integrate ChatGPT-powered conversational UI directly into their own products.

Codex Updates & SDK

OpenAI didn't leave its coding assistant behind: Codex got several upgrades - including enterprise controls, a Slack integration, and a Codex SDK to treat code generation as a pluggable component.

Platform & Ecosystem Moves

ChatGPT's Internal "App Store"

One of the biggest shifts was turning ChatGPT into a platform with first-class "apps." OpenAI plans to accept app submissions later in 2025, provide monetization (commerce) features, and launch an app directory where high-quality apps get higher visibility.

Initial app partners include Canva, Coursera, Figma, Booking.com, Spotify, Zillow, Expedia and more.

Metrics & Scale

At DevDay, Altman revealed that OpenAI's ecosystem now supports 800 million weekly ChatGPT users and 4 million developers. The API processes around 6 to 8 billion tokens per minute (depending on the source), indicating the sheer compute scale behind the scenes.

OpenAI also affirmed that growth and product innovation still take precedence over short-term profitability.

Infrastructure & Partnerships

AMD & Compute Strategy

OpenAI confirmed a multi-year partnership with AMD to supply high-performance chips, particularly the upcoming MI450, to support future compute expansion. Under the deal, OpenAI may acquire up to 160 million AMD shares (~10%), tied to performance milestones.

The ambition is to scale toward 6 gigawatts of compute capacity (starting with 1 GW), broadening beyond its reliance on Nvidia.

Altman and Brockman admitted that compute remains one of the most binding constraints on AI development.

Enterprise & Strategic Partnerships

OpenAI made clear its intent to accelerate enterprise growth, with new integrations and tie-ups (e.g. with Spotify, Zillow, Mattel) to embed AI across domains.

For example, the deal with Mattel was cited in conjunction with Sora's use in toy-concept generation.

Hardware Vision (with Jony Ive)

In a rarely publicized pairing, Sam Altman and Jony Ive discussed OpenAI's future hardware ambitions. Ive said the team is exploring 15-20 AI device ideas, aiming for technology that cultivates joy, connection, and emotional balance rather than pure productivity.

One concept under discussion is a screenless, voice-interactive companion device, roughly smartphone-sized, always listening, with built-in cameras and sensors. It would not be about flashy features but subtle ambient intelligence.

Ive emphasized that innovation must remain user-centered and humanistic, a counterweight to the fast pace of AI development.

Additional Announcements / Missed Highlights

• Open-weight model roadmap: Earlier in 2025, OpenAI had committed to releasing an open-weight reasoning model to give developers parameter access, enabling fine-tuning, a move toward more open AI ecosystems.
• Enterprise controls & security guardrails: New features announced include data retention policies, "disable training" toggles, tenancy isolation, audit trails, and private network/VPC support to satisfy compliance requirements.
• Operator & agent deprecation: The earlier "Operator" agent (browser automation) is being phased out in favor of the more general AgentKit approach.
• Pricing, throttle & throughput details: DevOps and procurement teams were asked to watch for live-streamed tables of token pricing, concurrency limits, SLOs and performance benchmarks that will shape enterprise adoption.

Implications & Outlook

DevDay 2025 reflects OpenAI's pivot from purely model-centric to platform-centric thinking. By embedding "apps" in ChatGPT, offering deep agent tooling, and expanding multimodal capabilities, the company is positioning itself as an operating layer for AI experiences.

The compute partnerships and hardware hints suggest OpenAI isn't content to be a software layer - it wants control over the full stack. Meanwhile, the volume of announcements underscores a growing challenge: staying coherent in product vision as the feature matrix explodes.

If OpenAI delivers on stability, scalability, and developer economics, DevDay 2025 may mark the moment it shifts from AI service vendor to AI platform company in earnest.