Wow, GPT5.4 has finally arrived! Several key points: 1. Native computer operation The Agent model that can directly "understand the screen" and operate the computer, supports mouse, keyboard, and cross application task execution, is truly capable of working. 2. Extra long context+stronger visual appeal Supports 1 million token context and can handle higher precision images, suitable for long process and multi-step tasks. 3. Stronger knowledge work Create tables PPT、 Documents perform better in this type of work. And there are fewer hallucinations: compared to GPT-5.2, complete error replies are reduced by 18%, and the probability of a single false message is reduced by 33%. 4. Continue to strengthen programming skills Inheriting the code capabilities of GPT-5.3-COdex, it has stronger capabilities in complex tasks, front-end development, and debugging. Codex has also added/fast mode, which can generate up to 1.5 times faster. 5. Smarter tool calling No longer cram all tool definitions into prompt words, but dynamically search when needed. The result is that token consumption can be reduced by up to 47%, and both speed and cost will be better. 6. Web search is more suitable for in-depth research Proficient in multi round, continuous, and cross source retrieval, suitable for complex information integration, not just simple searching for answers. 7. Stronger controllability In ChatGPT, GPT-5.4 Thinking will provide a preview of the thought process first, and users can even change direction directly during its response without having to restart a round. 8. It is currently being fully launched GPT-5.4 has started to enter ChatGPT, API, and Codex, gradually replacing the previous GPT-5.2 Thinking. Although the API unit price is higher, it saves more tokens, so the total cost may not necessarily be more expensive.