GPT-5.4 Launched in ChatGPT, brings major improvements today. OpenAI released the model with powerful new features. The system now handles complex digital work better.
The AI can operate apps and devices directly. This release focuses on automation and knowledge work. It also improves reasoning, coding, and factual accuracy. The update moves AI closer to autonomous agents.
Related Article: ChatGPT Starts Rolling Out Ads Officially in the U.S.
Key Takeaways:
- GPT-5.4 can control apps and computers directly.
- The model improves coding, research, and workflows.
- Benchmarks show strong accuracy and knowledge performance.
GPT-5.4 Launched in ChatGPT, Adds Computer Use
The AI can now operate devices directly. It works across many apps and software tools. Users can automate several digital tasks easily. The system handles spreadsheets, documents, and presentations smoothly.
The model can also write special control code. This code allows it to operate computers automatically. It can send keyboard and mouse commands. These actions respond to screenshots from the screen.
This feature helps the AI understand tasks visually. It can see screens and react correctly. The system performs actions across many applications. This helps complete workflows inside different programs.
The AI also improved web browsing performance. It interacts better with tools and APIs. These abilities improve automation across digital platforms.
The release also supports agentic AI systems. Agentic systems allow AI to complete tasks automatically. Multiple AI agents may work together silently. They finish complex tasks in the background.
OpenAI earlier introduced a tool called ChatGPT Agent. That system can control computers directly. GPT 5.4 performs tasks like searching for information online. It may also purchase items through websites. It continues progress toward this AI future.
Related Article: How to Fix a Deactivated ChatGPT Account
GPT-5.4 Improves Knowledge Work and Research
GPT-5.4 greatly improves professional knowledge work. The system combines reasoning and coding abilities. It also includes advanced agentic workflow systems.
The model gained coding power from GPT-5.3 Codex. This improves programming and development tasks significantly. Developers can write and test code faster.
The system also creates professional work documents. It generates presentations quickly and clearly. It also builds spreadsheets and structured reports. These features help many business professionals daily.
Researchers may also benefit from better reasoning. The AI answers complex questions more accurately.
The model can search the web more deeply. It finds information that is difficult to find online. The AI gathers information from many sources. Then it combines them into clear answers.
The Codex application supports a very large context. It can handle up to one million tokens. This helps manage large documents and datasets.
Related Article: How to Browse the Web with ChatGPT Atlas
GPT-5.4 Shows Strong Benchmark Performance
GPT-5.4 also achieved strong benchmark performance scores. Many tests measure knowledge and coding skills.
The GDPval benchmark measures professional knowledge work. GPT-5.4 scored eighty-three percent in this test. The benchmark evaluates forty-four real occupations.
Another model scored slightly lower results. Claude Opus 4.6 achieved seventy-eight percent.
Coding ability improved in SWE Bench Pro testing. GPT-5.4 scored fifty-seven point seven percent there. This benchmark measures real coding problem-solving.
Computer control performance also improved significantly. OSWorld Verified benchmark tests computer operation tasks. GPT-5.4 achieved seventy-five percent accuracy here.
Scientific knowledge testing also showed strong performance. GPQA Diamond measures advanced science questions. GPT-5.4 scored ninety-two point eight percent accuracy.
Another AI model scored slightly higher there. Gemini 3.1 Pro achieved ninety-four point three percent.
OpenAI also reduced false information significantly. The model produces fewer incorrect statements now. False claims dropped by thirty-three percent. Errors also reduced compared with GPT-5.2. The model now shows eighteen percent fewer mistakes.
Related Article: ChatGPT Pushes Health Feature and How you can enter the waitlist
Availability and Pricing
OpenAI released the model across several platforms. GPT-5.4 is available through the OpenAI API. Developers can access the model for applications.
The company also supports the Codex coding assistant. Codex helps developers write and manage software.
A reasoning version called GPT-5.4 Thinking exists. This version focuses on deeper reasoning tasks.
GPT-5.4 Thinking is coming to ChatGPT users. ChatGPT Plus and Team users will receive access. ChatGPT Pro users will receive GPT-5.4 Pro. This version offers stronger performance capabilities.
OpenAI also shared the API pricing details. Input tokens cost two dollars and fifty cents. Output tokens cost fifteen dollars per million.
The End Note
GPT-5.4 Launched in ChatGPT, represents important AI progress. The model introduces powerful computer use abilities. It can control software and complete tasks automatically.
These improvements move AI toward autonomous digital assistants. Future systems may complete many tasks independently.
For the latest tech news, follow Hogatoga on Twitter, Facebook, and Google News For the latest tech-related videos, Subscribe to our YouTube Channel and Newsletter.
Thanks