GPT 5.4 Unveiled: OpenAI's Latest AI Model Redefines Developer Capabilities

OpenAI has rolled out GPT 5.4, heralding it as the “best AI model ever made,” with a particular emphasis on its benefits for software developers. The new iteration integrates advanced reasoning, coding, and agentic workflow capabilities, marking a substantial evolution from its predecessors. Key improvements include enhanced steerability, significantly better context handling with support for up to one million tokens, and increased token efficiency, particularly for complex reasoning tasks. The model’s improved ability to process user interruptions mid-reasoning and its focus on programmatic interactions for browser usage, including explicit training to run JavaScript for UI automation, position it as a powerful tool for complex development tasks. OpenAI has also clarified its model nomenclature, introducing 5.4 Thinking and 5.4 Pro, while implying a potential deprecation of dedicated ‘Codex’ model variants as their functionalities are absorbed into the core architecture.

Performance benchmarks reveal GPT 5.4’s state-of-the-art status. On internal metrics like SBench Pro, it achieved a score of 57.7, surpassing previous records. Proprietary benchmarks, such as Skatebench V2, show 5.4 High leading at 82%, with 5.4 X-High at 81% and 5.4 Pro at 79%, demonstrating varied performance across its tiers. Notably, 5.4 Pro, despite its higher cost, has shown to underperform standard 5.4 in some benchmarks, though it has uniquely solved highly challenging problems like Defcon’s Goldbug Cshanty puzzle. Pricing for GPT 5.4 has increased to $2.50 per million input tokens and $15 per million output tokens, while 5.4 Pro commands a significantly higher $30 per million input and $180 per million output, reflecting a presumed increase in operational costs. However, the model exhibits a regression in prompt injection vulnerability during function calls, failing 2% of the time in specific tests, a concern for applications relying on external data. Despite its strengths, GPT 5.4 continues to lag behind models like Opus and Gemini in UI and frontend design, a sentiment echoed by expert testers. GPT 5.4 is currently available via ChatGPT (5.4 Thinking) and integrated into third-party platforms like T3 Chat, with forthcoming support in development environments like T3 Code.