The newly upgraded GPT 5.4 and 5.4pro are really amazing, you’ve got to try them out and experience it for yourself. The official announcement says GPT-5.4 is one of their first general models equipped with cutting-edge native computer use capabilities. It achieved 75.0% on OSWorld-Verified, higher than GPT-5.2’s 47.3% and even surpassing the human benchmark of 72.4%. The release notes also mentioned improvements in tool selection, cross-software environment workflows, and deep web research. This means 5.4 is better suited for long-chain reasoning, excels at deep web research, and integrates multiple sources more effectively than before. After using it for a day, my biggest takeaway is that there are absolutely no typos in Chinese anymore. Previously, whether it was images or PDFs from Shenzhen, there were always tons of Chinese typos. Also, its capabilities have seen qualitative improvements in many areas, like reasoning, research, summarization, charting, and table creation. Plus, in multi-step analysis, the conclusions and structure are much more stable. It’s absolutely incredible. I feel like I probably won’t be opening Claude or Gemini much anytime soon. Of course, Gemini is still the most comfortable for practicing English. For now, GPT hasn’t completely dominated other functionalities, but it’s definitely taken a big step forward.