OpenAI has introduced its advanced model, GPT-5.4, across ChatGPT, Codex, and its API, enhancing reasoning and agent capabilities for professional applications. Released as GPT-5.4 and GPT-5.4 Pro, the model effectively supports over 40 professions, including finance and law, and minimizes the need for separate coding models. It boasts a 75% success rate on the OSWorld-Verified benchmark, surpassing both GPT-5.2 and human benchmarks. The model excels in web navigation and automates workflows using screenshots, mouse clicks, and keyboard inputs, making it suitable for tasks like document editing and spreadsheet management. With improved reasoning transparency and reduced false claims, GPT-5.4 sets a high bar in professional service tools. Priced competitively at $2.50 per million input tokens, it challenges competitors like Claude Opus 4.6 while blending extensive capabilities and cost efficiency. OpenAI’s new tool-search method lessens token usage, enhancing long-context capabilities of up to 1.05 million tokens.
Source link
