In the wake of escalating discussions and controversies surrounding a pivotal contract with the U.S. Department of Defense, OpenAI has unveiled its latest model, GPT-5.4. This release emerges at a crucial time when users have been increasingly migrating from OpenAI’s flagship ChatGPT to competing platforms, particularly Anthropic’s Claude, indicating a shifting landscape in the artificial intelligence sector.
The rollout of GPT-5.4 is being implemented gradually across the ChatGPT service and Codex, with availability through the API labeled as gpt-5.4. Users who subscribe to Plus, Team, and Pro plans can access GPT-5.4 Thinking in ChatGPT, while Enterprise and educational customers have the option to enable early access. Furthermore, GPT-5.4 Pro can be accessed via the API as gpt-5.4-pro and is available to both Pro and Enterprise plans.
In its official announcement, OpenAI characterized GPT-5.4 as a flagship model that encapsulates the company’s latest advancements in areas such as reasoning, coding, and agent-based workflows. The organization articulated that this new model builds upon its predecessor, GPT-5.3-Codex, and significantly enhances capabilities in managing tools, software environments, and various routine office tasks.
OpenAI claims that GPT-5.4 represents a marked improvement in handling professional tasks such as spreadsheet management, presentation creation, and document processing. Additionally, the new model boasts enhanced coding capabilities, support for intricate multi-step tasks, a larger context window, and improved factual accuracy, producing fewer hallucinations compared to earlier iterations.
“We have dedicated our efforts to making GPT-5.4 more adept at real-world applications by reducing hallucinations and errors. This model is the most factual we’ve released so far. When tested against a dataset of de-identified prompts, GPT-5.4 exhibited a 33% reduction in the likelihood of individual claims being false, and an 18% decrease in full responses containing any errors, compared to GPT-5.2,” OpenAI noted.
Moreover, the tech company asserts that GPT-5.4 is its first general-purpose model equipped with native computer-use capabilities. OpenAI views this development as a significant advancement for both developers and AI agents, indicating that the model is designed to navigate complex tasks with greater efficiency.
Enhanced Safety Measures
In conjunction with the introduction of GPT-5.4, OpenAI disclosed enhancements made to safety protocols. The company has bolstered the safeguards initially in place for GPT-5.3-Codex, maintaining a high cyber-risk classification while implementing additional protective measures as detailed in its system documentation. These measures encompass expanded cyber safety systems, advanced monitoring tools, trusted access controls, and the implementation of request blocking to mitigate higher-risk activities on Zero Data Retention surfaces.
OpenAI describes its precautionary approach as essential, particularly due to the dual-use capabilities inherent in cybersecurity technologies. The firm has acknowledged the potential for detection systems to yield false positives, a risk that they are actively working to mitigate. Ultimately, the emphasis is on reducing unnecessary refusals while ensuring that strong protections remain against potential misuse of the model.
Additionally, OpenAI has made strides in safety research concerning the reasoning capabilities of its models. They recently released an open-source evaluation designed to assess whether these systems can effectively conceal their reasoning processes. Findings indicated that GPT-5.4 Thinking displayed a low capacity for obscuring reasoning, a characteristic that OpenAI has framed as a favorable safety indicator.
As artificial intelligence continues to evolve and shape the digital landscape, the release of GPT-5.4 underscores OpenAI’s commitment to improving user experience while addressing critical safety concerns. This dual focus on advanced capabilities and robust protective measures illustrates OpenAI’s strategic approach to navigating the complex challenges that lie ahead in the ever-competitive field of AI technology. As the company seeks to regain user trust and stand out amidst rising competition, the implications of GPT-5.4 extend beyond mere technological advancements, resonating within the broader discourse surrounding responsible AI development.