GPT-5.4 Arrives in ChatGPT with Native Computer Use and Agentic Capabilities

openai launched gpt-5.4 thinking
Image Credit: OpenAI
In Short
  • OpenAI has unveiled GPT-5.4, its frontier AI model for computer use, agentic workflows, web research, and professional knowledge work.
  • The new model supports native computer use capabilities and it can operate apps and websites much better than previous models.
  • GPT-5.4 Thinking is already rolling out in ChatGPT and GPT-5.4 Pro is coming to ChatGPT Pro users, starting today.

OpenAI has launched its next-gen GPT-5.4 and GPT-5.4 Pro models. The new release focuses on improving professional knowledge work, computer use, agentic capabilities, and better token efficiency. The GPT-5.4 in ChatGPT Thinking model delivers state-of-the-art performance in computer use as it can operate apps and execute multi-step workflows. Its tool search capability has also improved dramatically.

GPT-5.4 Improves Agentic Workflow, Research, and Knowledge Work

The new model integrates advanced reasoning, coding prowess from GPT-5.3-Codex, and agentic workflows into a single model. It’s also much better at generating presentations, spreadsheets, and documents for professional knowledge work. It also supports context window of up to 1 million tokens in the Codex app.

Coming to benchmarks, the new GPT-5.4 model achieved 83% in GDPval, which tests professional knowledge work across 44 occupations. For context, Anthropic’s Claude Opus 4.6 got 78% in GDPval. Next, in SWE-Bench Pro which evaluates coding performance, GPT-5.4 scored 57.7% and in OSWorld-Verified benchmark which tests computer use, the model achieved 75% accuracy.

Image Credit: OpenAI

Next, GPT-5.4 is better at agentic web browsing, and it means that the model can find hard-to-locate information on the web. As for scientific knowledge, GPT-5.4 scored 92.8% in GPQA Diamond while Google’s Gemini 3.1 Pro did slightly better, achieving 94.3%.

Finally, OpenAI has managed to reduce hallucinations on GPT-5.4. According to the company, it makes 33% fewer false claims and compared to GPT-5.2, the new model has 18% fewer errors.

OpenAI says GPT-5.4 Thinking is rolling out to ChatGPT Plus, and Team users. ChatGPT Pro users will get the more powerful GPT-5.4 Pro model. As for API pricing, GPT-5.4 costs $2.50 for one million input tokens and $15 for one million output tokens.

Comments 0
Leave a Reply

Loading comments...