OpenAI
APIs and tools for building AI products
5.0•696 reviews•11K followers
APIs and tools for building AI products
5.0•696 reviews•11K followers
The most powerful platform for building AI products.
Build and scale AI experiences powered by industry-leading models and tools.
This is the 25th launch from OpenAI. View more
GPT‑5.4
Launching today
GPT-5.4 Thinking delivers deeper web research, stronger context retention on long tasks, and 33% fewer factual errors than its predecessor. You can now interrupt the model mid-response and redirect it. No need to start over. Same intelligence. More control. Less token burn by default.
Free
Launch Team
Gemini 2.5 Pro
Excited to hunt GPT-5.4 today!
This is OpenAI's most capable reasoning model yet and it's not just an incremental bump. GPT-5.4 merges the coding power of GPT-5.3-Codex with serious knowledge work and native computer-use capabilities into one model. Less back and forth, more actual output.
What stands out:
-Native computer use: the model can operate a desktop, click, type, navigate apps
-Matches or beats industry professionals on 83% of real-world knowledge tasks (GDPval)
-33% fewer factual errors compared to GPT-5.2
-Tool search cuts token usage by 47% in large tool ecosystems
-1M context window support in Codex
-Significantly better at spreadsheets, presentations, and documents
It's not trying to wow you with a feature list. It's trying to actually finish the work you give it. Faster, with fewer mistakes, and with less hand-holding.
The computer use benchmark result alone (75% on OSWorld-Verified, surpassing human performance at 72.4%) is the kind of number that makes you stop and think.
Follow me on Product Hunt to stay on top of the biggest launches in AI: @byalexai
Impressive numbers! Though benchmarking against your own previous models is a bit like winning a race you organized, against yourself. Would love to see how it stacks up against the rest of the field. Either way, excited to try it in Codex!
BlocPad - Project & Team Workspace
The mid-response interruption feature is honestly what I've been waiting for. So many times I realize halfway through a response that I asked the wrong thing and just have to sit there watching tokens burn. 33% fewer factual errors is a big claim too, curious how that holds up on more niche technical domains.
Built my entire product, Fillix, an AI job application automation tool, on OpenAI's API. The reliability and speed of the models is what makes real-time form-filling actually viable. Structured outputs changed the game for us. Keep shipping
The depth of APIs and tooling here gives builders a lot to work with. What kinds of AI products are you most excited to see launched next on this stack?
git-lrc
Two models in the span of 24-48 hours, crazy!