|
|
A lot has shipped lately. We’ve rolled out updates across models, Codex, and APIs, and developers are already turning these releases into new products, games, workflows, and tools.
But if we had to pick the thing we’re most excited about right now 🥁….. DevDay 2026 is officially on the calendar! Save the date and enter our contest for a chance to win a ticket.
To get live answers to your lingering product questions in the meantime, join our upcoming online Build Hours:
|
|
|
|
|
What shipped, what changed, what to automate next
|
A new wave of voice intelligence
We’re excited to announce three new models in the Realtime API that help you build smarter voice agents that feel more natural in conversation and can take action live as speech is happening:
- gpt-realtime-2: Smarter voice agents with stronger instruction following, prompt adherence, and multilingual capabilities
- gpt-realtime-translate: Translate conversations live across 70 input languages and 13 output languages
- gpt-realtime-whisper: Transcribe each word as its spoken rather than waiting for a sentence to finish
It’s one of those things you have to see (and hear) for yourself. Check it out in action and try it in the playground.
|
Codex can now use Chrome directly
So much work happens in browser tabs. The new Chrome extension builds on computer use, so Codex can work directly inside sites where you’re signed in, organize task-specific tabs, and hand back results for review.
|
|
|
Other updates:
- GPT-5.5 is our most intelligent model yet with stronger reasoning, improved tool use, and more reliable long-running execution.
- Agents SDK lets agents inspect files, run commands, edit code, and work on long-horizon tasks within controlled environments. It's powered by a customizable harness and lets you bring-your-own sandbox from Modal, E2B, Daytona, Cloudflare, and more.
- Codex for (almost) anything! Add background computer use, browser interaction, and 90+ new plugins to accelerate your workflows.
- GPT-Image-2 lets you design production-quality images.
Here are a couple favorites from the showcase gallery:
|
|
|
Watchmaker Landing Page
A polished storefront demo that turns a concept into a branded product experience fast with GPT-5.5 and image generation.
|
|
|
Swifty Dungeon Game
A native macOS game built with Codex, GPT-5.5, and generated visual assets.
|
|
|
|
|
|
|
Guides, blogs, and things worth cloning
|
Migrate a legacy codebase with sandbox agents: A code-migration agent should work in a controlled environment, one scoped task at a time: inspect the relevant repo, edit files, run checks, and return a patch. The cookbook shows agents doing real repo work, not just chatting about code.
Unleash your agents in the docs: OpenAI's public docs MCP server lets agents search and read official docs from your editor or CLI. It plugs directly into Codex, Cursor, and VS Code agent workflows to build doc-aware helpers and custom guides.
Prompt GPT-5.5 around outcomes, not steps: Use the GPT-5.5 prompting guide to define goals, constraints, and desired outputs. Then use our docs skill to apply the recommendations with Codex, and this migration guide to audit prompts, tool calls, and workflow reliability.
Prompt better in Codex: This is the best read-this-before-you-prompt-blindly resource for teams using Codex directly through the API.
API deployment checklist: If you’re moving to production with the Responses API, check out this guide to the highest-leverage design choices. This includes reasoning effort, tool_search, built-in tools, compaction, prompt caching, background mode, and if you’re getting fancy with it, WebSocket mode to speed up agentic runs with 20+ tool calls by 20-40%.
Codex use cases: Try using Codex in new ways with ideas and prompts from the Codex use case library. We now have 48 defined use cases and counting!
|
|
|
|
|
Go behind the scenes with OpenAI engineers
|
Speeding up agentic workflows with WebSockets in the Responses API
Hear from Brian Yu and Ashwin Nathan on turning model speed gains into visible workflow speed.
“It's been incredible seeing the advancements in inference systems and watching the latency bottleneck shift from model inference to infrastructure. We wanted to share those lessons with developers building the next wave of fast, agentic products.”
Symphony: orchestrating Codex at ticket level
Alex Kotliarskyi, Victor Zhu, and Zach Brock explain how Symphony turns an issue tracker into a control plane for coding agents, driving a 500% increase in landed pull requests.
“Six months ago we made a pretty wild decision: no human-written code in the repo. That constraint forced us to rethink how eng teams and agents work together. Symphony came out of the realization that human context was the limiting factor.”
How OpenAI delivers low-latency voice AI at scale
Yi Zhang and William McDonald outline how OpenAI rebuilt its WebRTC infrastructure with a split relay and transceiver architecture to deliver low-latency voice AI at scale without changing the client-side WebRTC experience.
“Building real-time voice infrastructure at scale forces you to care about every millisecond in a very human way. Writing this piece was a chance to share the engineering decisions that helped conversations with AI start feeling less like software and more like natural dialogue.”
|
|
|
|
|
From side projects to professional work
|
|
|
|
|
From driving solar-powered electric cars to open-sourcing AI prototypes, Ashe Magelhaes shares how she’s weaving AI into her creative process, product workflow, and everyday life.
“The name of the game is speed. You could build so much software nowadays, but what will you build that you yourself use every day, that you find to be a beautiful and elegant experience, and that you can surface to users in a way that’s sustainable?”
Happy building and vibing,
The OpenAI Team
|
|
|
|
|
|
|
|
|
Комментариев нет:
Отправить комментарий