Technical

Srinivasa Reddy Kandi: OpenAI Introduces GPT-5.4 With Pro and Thinking Variants for Advanced Professional Tasks

March, 06, 2026-04:11

Share: Facebook | Twitter | Whatsapp | Linkedin | Visits: 38319 | :2821

---

OpenAI Introduces GPT-5.4 With Pro and Thinking Variants for Advanced Professional Tasks:

OpenAI has released GPT-5.4, its latest foundation model designed to deliver stronger performance and greater efficiency for professional applications. The new model is offered in multiple versions, including the standard GPT-5.4, a reasoning-focused version called GPT-5.4 Thinking, and a high-performance edition known as GPT-5.4 Pro.

The API version of GPT-5.4 supports context windows of up to 1 million tokens, representing the largest context capacity currently provided by the company. This expanded capability allows the model to analyze and process significantly larger documents or datasets in a single request.

OpenAI highlighted improvements in token efficiency, noting that GPT-5.4 can complete many tasks using far fewer tokens compared with its predecessor, GPT-5.2. This efficiency could lower costs and speed up workflows for developers and organizations using the system.

In benchmark testing, GPT-5.4 demonstrated notable performance gains. It achieved record results on computer-use benchmarks such as OSWorld-Verified and WebArena Verified. The model also scored 83% on OpenAI’s GDPval benchmark, which evaluates AI performance in knowledge-intensive professional tasks.

Additionally, GPT-5.4 ranked highest on the APEX-Agents Benchmark, developed by Mercor. According to Mercor CEO Brendan Foody, the model performs particularly well at producing complex outputs such as slide presentations, financial models, and legal analysis while maintaining faster performance and lower operational costs than competing systems.

OpenAI also reported improvements in reliability. The company said GPT-5.4 is 33% less likely to make incorrect individual claims compared with GPT-5.2, and overall responses are 18% less likely to contain factual errors, reflecting ongoing efforts to reduce hallucinations and improve accuracy in AI-generated content.

Author: Kandi Srinivasa Reddy, Srinivasa Reddy Kandi, #KandiSrinivasaReddy, #SrinivasaReddyKandi