OpenAI Launches GPT-5.4 With Native Computer Use and Financial Plugins as It Fights to Regain User Trust
OpenAI releases its most capable model yet with autonomous computer operation and enterprise integrations, but the launch is overshadowed by Pentagon deal fallout and a reported 1.5 million user exodus.
Overview
OpenAI released GPT-5.4 on March 5, 2026, billing it as its most capable and efficient frontier model for professional work. The model arrives in three variants — standard GPT-5.4, GPT-5.4 Thinking for extended reasoning, and GPT-5.4 Pro for maximum performance — and introduces native computer-use capabilities alongside new enterprise integrations for financial workflows. The launch comes at a turbulent moment for the company, which has been contending with user defections and internal dissent following its controversial Department of Defense partnership.
What We Know
Capabilities and Performance
GPT-5.4 consolidates previously separate model capabilities into a single system, combining the advanced reasoning of prior GPT-5 variants with the coding strengths inherited from GPT-5.3 Codex. The API version supports context windows of up to one million tokens, the largest OpenAI has offered to date.
OpenAI claims the model is 33 percent less likely to produce false individual claims compared to GPT-5.2, with full responses 18 percent less likely to contain any errors. The company also emphasized improved token efficiency, stating that GPT-5.4 can solve comparable problems while consuming significantly fewer tokens than its predecessors.
Native Computer Use
The standout technical addition is out-of-the-box autonomous computer operation. GPT-5.4 can operate computers and software autonomously, search for and use external tools on demand, and handle complex multi-step tasks without requiring developers to build supporting infrastructure. The model can execute keyboard and mouse commands and navigate operating systems directly, positioning OpenAI alongside competitors in the growing agentic AI market.
Enterprise and Financial Integrations
OpenAI is pairing the model release with new enterprise tools, including ChatGPT integrations for Microsoft Excel and Google Sheets currently in beta. Financial data plug-ins from FactSet, MSCI, Third Bridge, and Moody’s enable teams to pull market, company, and internal data into a single workflow. The financial services push directly challenges Anthropic, which launched similar products in July 2025.
Safety Measures
GPT-5.4 retains a high cybersecurity capability classification under OpenAI’s Preparedness Framework and ships with a two-tiered monitoring architecture that uses a topical classifier to identify cybersecurity-related content and a safety reasoner to categorize responses within a threat taxonomy. The company’s Trusted Access for Cyber program gates high-risk dual-use capabilities behind an identity-based enrollment system for enterprise defenders. OpenAI also published an open-source evaluation called CoT controllability, designed to detect whether reasoning models can deliberately obscure their chain-of-thought. The system card reports that GPT-5.4 Thinking successfully controlled only 0.3 percent of outputs on tasks with 10,000-character reasoning traces, which the company characterized as a positive safety signal indicating limited ability to hide its reasoning from monitors.
Availability
GPT-5.4 Thinking is available to ChatGPT Plus, Team, and Pro users, while GPT-5.4 Pro is limited to Pro and Enterprise plans. API access is available across tiers, with per-token costs slightly higher than GPT-5.2, though OpenAI argues the efficiency gains offset the increase.
What We Don’t Know
OpenAI has not disclosed full benchmark comparisons against competing models from Anthropic, Google DeepMind, or Chinese labs. The company’s claim that GPT-5.4 tops Mercor’s APEX-Agents leaderboard for professional services work has not been independently verified. It also remains unclear how the computer-use feature will be monitored in practice, particularly given the high cyber-risk classification and the potential for autonomous actions to produce unintended consequences at scale.
Analysis
The timing of GPT-5.4’s release is difficult to separate from OpenAI’s broader reputational challenges. The company reportedly lost approximately 1.5 million users after announcing its Pentagon partnership, with some employees openly opposing the decision and CEO Sam Altman acknowledging the backlash was “really painful.” Whether a technically stronger model can reverse that trajectory remains an open question. The enterprise-focused financial integrations suggest OpenAI is prioritizing revenue diversification at a moment when its consumer brand has taken a hit, while the safety disclosures appear calibrated to address concerns about the very capabilities — autonomous computer operation and agentic behavior — that the model is designed to showcase.