OpenAI Unveils GPT 5.4: A Revolutionary AI Model

OpenAI has introduced its latest model, GPT 5.4, which brings together enhanced reasoning, coding abilities, and superior performance in a unique flagship version. Alongside a standard edition, GPT-5.4 is offered in two specific variations: GPT-5.4 Thinking for improved reasoning and GPT-5.4 Pro for top-tier performance on intricate tasks.

What’s New in GPT-5.4

The GPT 5.4 Thinking variant in ChatGPT allows users to see its reasoning process prior to completing a response. This feature enables users to make corrections while the model is still generating its output, resulting in a more precise and cohesive final answer.

Furthermore, the API version of this model features an extensive context window of 1 million tokens. OpenAI indicates that this iteration includes computer-use functionalities, enabling agents to manage computers and execute complex workflows across multiple applications.

Notably, compared to GPT-5.2, GPT-5.4 uses fewer tokens to achieve similar conclusions, making it the most efficient reasoning model to date in both speed and cost.

OpenAI also asserts that GPT-5.4 ranks as the most accurate model thus far; individual statements are 33% less likely to be incorrect, while overall responses contain 18% fewer errors in comparison to GPT-5.2.

Outstanding Benchmarks

The GPT-5.4 garnered a score of 83% on OpenAI’s GDPval test, notably higher than the 70.9% achieved by GPT-5.2. This assessment evaluates the model’s efficacy in generating well-defined knowledge work across 44 professions. In internal benchmarks for spreadsheet modelling tasks, GPT-5.4 scored 87.3%, surpassing GPT-5.2’s score of 68.4%.

Brendan Foody, CEO of Mercor, noted that GPT-5.4 excels in producing long-term deliverables, such as slide presentations, financial models, and legal analyses, while outperforming rival models in terms of speed and cost efficiently.

The model also achieved impressive results in computer usage benchmarks, certified by OSWorld and WebArena, solidifying its position as one of the company’s most advanced AI models. In terms of coding, GPT 5.4 reportedly provides up to 1.5 times faster token velocity, allowing for swift progress through coding tasks, iterations, and debugging while maintaining a seamless workflow.

Availability and Pricing

In ChatGPT, the GPT-5.4 Thinking model is accessible to ChatGPT Plus, Team, and Pro users. The GPT-5.2 Thinking model will remain available for additional three months for paid subscribers before it is discontinued on June 5, 2026.

For Pro and Enterprise users, GPT-5.4 Pro is accessible.

In the API, pricing for GPT-5.4 is set at $2.50 for every million input tokens and $15 for every million output tokens.

On the other hand, GPT-5.4 Pro is priced at $30 for every million input tokens and $180 for every million output tokens.