GPT 5.4 arrives on ChatGPT: 5 improvements to know

0
963

GPT 5.4 arrives on ChatGPT: 5 improvements to know

OpenAI dropped GPT‑5.4 this week, and the company said ChatGPT power users should notice improvements right away. The launch comes just days after the launch of GPT-5.3 Instant, a more conversational model geared toward everyday users.

OpenAI wrote that GPT-5.4 is the "most capable and efficient frontier model for professional work." So, obviously, GPT-5.4 is designed for professionals and developers. The company also announced the launch of GPT-5.4 Thinking and GPT‑5.4 Pro, aimed at people who want peak performance.

Here are three improvements that OpenAI made.

Improved efficiency

OpenAI said that GPT-5.4 operates more efficiently than ever.

It wrote that it was the company's "most token-efficient reasoning model yet, using significantly fewer tokens to solve problems when compared to GPT‑5.2, translating to reduced token usage and faster speeds." Basically, that means GPT-5.4 can solve problems and deliver answers with fewer tokens.

Better at doing real-world tasks

OpenAI said that the new model was far more capable of completing real-world tasks related to knowledge work.

Wrote OpenAI: "On GDPval⁠, which tests agents’ abilities to produce well-specified knowledge work across 44 occupations, GPT‑5.4 achieves a new state of the art, matching or exceeding industry professionals in 83.0% of comparisons, compared to 70.9% for GPT‑5.2."

Mashable Light Speed

In other words, it's capable of doing a lot of people's jobs reasonably well.

Improved accuracy

OpenAI claimed that GPT-5.4 cut down on mistakes. Compared with GPT-5.2, OpenAI said the model is 33 percent less likely to make false claims and its claims were 18 percent less likely to contain any errors, according to OpenAI.

More flexible approach to thinking

Instead of simply giving GPT-5.4 a prompt and letting it do its thing, you can course-correct the model as it goes.

"In ChatGPT, GPT‑5.4 Thinking can now provide an upfront plan of its thinking, so you can adjust course mid-response while it’s working, and arrive at a final output that’s more closely aligned with what you need without additional turns."

ChatGPT for Excel add-in

Enterprise customers can also use GPT-5.4 with a new OpenAI plugin, the ChatGPT for Excel add-in. per OpenAI, the latest models will be more capable of creating and analyzing spreadsheets.

How to try GPT-5.4

As per usual, the new model starts rolling out today. It will be available in ChatGPT, Codex, and via the API. For this release, you'll need a paid ChatGPT subscription to use the new models, at least, for now.

You can read OpenAI's full blog post about the launch at the company's website.


Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.

Zoeken
Categorieën
Read More
Technology
Score a 2025 Mini-LED TV for $530 at Amazon
65-inch TCL QM6K Series Smart Google TV: $529.99...
By Test Blogger7 2026-02-17 00:00:14 0 1K
Spellen
Ubisoft axes six games, including the Prince of Persia Sands of Time remake that felt like it was finally within reach
Ubisoft axes six games, including the Prince of Persia Sands of Time remake that felt like it was...
By Test Blogger6 2026-01-23 23:00:35 0 2K
Home & Garden
Christmas or Easter Cactus? How to Spot the Difference to Help Yours Thrive
Not Sure Which Holiday Cactus You Have? Here’s How to Quickly Spot the Difference Key Takeaways...
By Test Blogger9 2026-03-16 19:00:37 0 954
Spellen
"We'll keep making Warframe if there's one player count," Digital Extremes' Megan Everett asserts. "I don't care what SteamDB says"
"We'll keep making Warframe if there's one player count," Digital Extremes' Megan Everett...
By Test Blogger6 2026-03-28 10:00:16 0 757
Home & Garden
This Is the First Thing to Change If Your Bedroom Doesn't Feel Relaxing
Why Your Bedroom Doesn’t Feel Restful and the First Thing Experts Say to Change Credit: David...
By Test Blogger9 2026-03-15 07:00:32 0 816