OpenAI DevDay 2025: What Just Changed for Builders

Seven Model Updates: More Power, More Options

OpenAI DevDay 2025 focused on more than just new tools. The event highlighted upgrades to the technology powering them. These new models offer developers greater control, improved performance, and lower costs for voice, image, video, and reasoning tasks.

1. GPT-5 Pro

GPT-5 Pro is now available in the API and is OpenAI’s most advanced model so far. It is designed for high-stakes use cases such as legal, financial, and medical fields, or any area where precision and deep reasoning are important. Startups like Cursor, Windsurf, and Vercel are already using it to change how software is built, from agent workflows to complete coding solutions.

2. gpt-realtime-mini

This compact voice model offers the same natural tone and responsiveness as its larger version, but at a much lower cost. It is 70% cheaper and built for real-time interaction, making voice-based AI more accessible for everyday use. Altman emphasized that voice will become a primary way to interact with AI, and this model moves us closer to that goal.

3. gpt-image-1-mini

This lightweight image generation model costs 80% less than the full version. It is ideal for production use, where speed and cost are important, but quality still matters.

4. Sora 2 API (Preview)

Developers now have access to the same video generation model that powers Sora 2. It’s a game-changer for creators, educators, designers, and filmmakers.

What sets it apart:

• You can guide the model using detailed prompts, and it keeps track of context from one frame to the next. This helps it create videos that look consistent, stylish, and visually appealing.
• You can start with a video clip from your phone and expand it into a cinematic wide shot. Or you can take a still image and generate a scene around it, such as turning a photo of your dog into a playful group adventure.
  
  

Sound integration: 

Sora 2 does more than generate visuals; it also pairs them with immersive audio. From ambient sounds to synchronized effects, the model creates a layered experience that feels realistic. In one demo, a kayaking scene included the splash of paddles and environmental sounds that matched the visuals perfectly.

Controls available:

• Video length
• Aspect ratio and resolution
• Easy remixing and editing
  
  

Real-world example:

Mattel is using Sora 2 to turn sketches into shareable product videos. Instead of static mockups, designers can show how toys move, how kids might interact with them, and share those ideas with stakeholders more quickly than with traditional prototyping.