OpenAI officially launched GPT-5.4, its most powerful and efficient “frontier model” to date. Positioned as a professional-grade upgrade, the model consolidates the previously fragmented GPT-5 lineup—unifying the coding power of GPT-5.3 Codex with the deep reasoning of the Thinking series into a single flagship system.
The release follows a period of intense brand volatility, specifically aiming to regain market confidence after the company’s recent controversial defense deal.
The Three Tiers of GPT-5.4
OpenAI has moved away from a “one-size-fits-all” approach, offering three distinct versions of the 5.4 architecture:
- GPT-5.4 Thinking: The default flagship for ChatGPT Plus, Team, and Business users. It is designed for deep research, complex reasoning, and long-horizon tasks.
- GPT-5.4 Pro: A “maximum performance” variant exclusive to ChatGPT Pro ($200/mo) and Enterprise users. It is optimized for “research-grade intelligence” on the most difficult math and engineering problems.
- GPT-5.4 (API/Codex): The developer-facing model, featuring a massive 1 million token context window, making it capable of analyzing entire codebases or hundreds of legal documents in a single prompt.
Key Breakthroughs & Features
GPT-5.4 introduces several “world-first” capabilities for a general-purpose model, shifting the AI from a chatbot to a digital agent.
- Native Computer Use: This is the first mainline model that can “operate” a computer. It can navigate software, click buttons, and type in applications by “looking” at screenshots. On the OSWorld-Verified benchmark, it scored 75%, outperforming the average human tester (72.4%).
- Upfront Reasoning Plans: In ChatGPT, GPT-5.4 Thinking now shows you a step-by-step plan of how it intends to solve a problem before it starts writing. You can interrupt and adjust the plan mid-stream to “steer” the model without restarting the conversation.
- 33% Fewer Factual Errors: OpenAI claims a significant reduction in hallucinations, with individual statements 33% less likely to be false compared to GPT-5.2.
- Built-in Tool Search: Instead of needing a list of tools provided by the developer, the model can now automatically search for and “load” the tools it needs to complete a task, significantly reducing API costs.
Benchmark Performance
OpenAI evaluated GPT-5.4 using GDPval, a test that measures an AI’s ability to perform the work of 44 different professional occupations.
| Benchmark | GPT-5.4 Score | Previous (GPT-5.2) |
| GDPval (Professional Tasks) | 83.0% | 70.9% |
| Spreadsheet Modeling | 87.3% | 68.4% |
| SWE-Bench Pro (Coding) | 57.7% | 55.6% |
| OSWorld (Computer Use) | 75.0% | 47.3% |
Pricing & Availability
- ChatGPT: GPT-5.4 Thinking is rolling out now to Plus, Team, and Pro users. It replaces GPT-5.2 Thinking, which will be moved to a “Legacy” section for three months.
- API Costs: The standard model is priced at $2.50 per 1M input tokens and $15.00 per 1M output tokens.
- Pro API: The Pro variant is significantly more expensive at $30.00 per 1M input tokens, targeting high-stakes enterprise applications.
