OpenAI Unveils GPT-5.4 Pro and Thinking Editions

On Thursday, OpenAI introduced GPT-5.4, a new foundation model described as "our most capable and efficient frontier model for professional work." Alongside the standard version, GPT-5.4 is offered as a reasoning-focused variant (GPT-5.4 Thinking) and a performance-optimized edition (GPT-5.4 Pro).
The API version of the model will support context windows of up to 1 million tokens, marking the largest context capacity OpenAI has ever offered.
OpenAI also highlighted enhanced token efficiency, noting that GPT-5.4 can solve identical problems using significantly fewer tokens than its predecessor.
The new model delivers substantially improved benchmark results, achieving record scores on the computer-use benchmarks OSWorld-Verified and WebArena Verified. It also set a new record with an 83% score on OpenAI's GDPval test for knowledge-work tasks.
According to a statement from Mercor CEO Brendan Foody, GPT-5.4 leads on Mercor's APEX-Agents benchmark, which evaluates professional skills in law and finance.
"[GPT-5.4] excels at producing long-horizon deliverables like slide decks, financial models, and legal analysis," Foody stated, "delivering top-tier performance while operating faster and at a lower cost than competing frontier models."
GPT-5.4 continues OpenAI's work to reduce hallucinations and factual inaccuracies. The company reports the new model is 33% less likely to make errors in individual claims compared to GPT-5.2, with overall responses 18% less likely to contain mistakes.
As part of the launch, OpenAI redesigned how the GPT-5.4 API handles tool calling, introducing a new system named Tool Search. Previously, system prompts had to define all available tools upfront—a process that consumed considerable tokens as tool libraries expanded. The new system lets models retrieve tool definitions on demand, making requests faster and more cost-effective in environments with many tools.
OpenAI has also added a new safety evaluation to assess its models' chain-of-thought—the running commentary that reveals the model's reasoning during multi-step tasks. AI safety researchers have long expressed concern that reasoning models might misrepresent their chain-of-thought, and testing confirms this can occur under certain conditions.
OpenAI's new evaluation indicates that deception is less probable in the Thinking version of GPT-5.4, "suggesting the model lacks the capability to conceal its reasoning and that CoT monitoring remains an effective safety tool."
Related article
Sam Altman Sparks Debate Over AI's Deceleration
Listen onApple PodcastsListen onSpotifyOpenAI CEO Sam Altman recently suggested that it may be time to “pace the rate of AI development” to allow society to “harden around some of these new capability levels.”On the latest episode of TechCrunch’s Equ
OpenAI fights Apple trade secret lawsuit
OpenAI rebutted Apple’s trade secret allegations on Tuesday, arguing the lawsuit is unfounded.“We take these claims seriously but see no evidence supporting them,” OpenAI stated, as reported by Bloomberg’s Ed Ludlow on X. “We support fair competition
OpenAI robotics head Caitlin Kalinowski resigns over Pentagon partnership
OpenAI robotics leader Caitlin Kalinowski has stepped down following the company’s controversial partnership with the Department of Defense.“This wasn’t an easy call,” Kalinowski explained in a social media statement. “While AI plays a vital role in
Related Special Topic Recommendations
Comments (0)
0/500

On Thursday, OpenAI introduced GPT-5.4, a new foundation model described as "our most capable and efficient frontier model for professional work." Alongside the standard version, GPT-5.4 is offered as a reasoning-focused variant (GPT-5.4 Thinking) and a performance-optimized edition (GPT-5.4 Pro).
The API version of the model will support context windows of up to 1 million tokens, marking the largest context capacity OpenAI has ever offered.
OpenAI also highlighted enhanced token efficiency, noting that GPT-5.4 can solve identical problems using significantly fewer tokens than its predecessor.
The new model delivers substantially improved benchmark results, achieving record scores on the computer-use benchmarks OSWorld-Verified and WebArena Verified. It also set a new record with an 83% score on OpenAI's GDPval test for knowledge-work tasks.
According to a statement from Mercor CEO Brendan Foody, GPT-5.4 leads on Mercor's APEX-Agents benchmark, which evaluates professional skills in law and finance.
"[GPT-5.4] excels at producing long-horizon deliverables like slide decks, financial models, and legal analysis," Foody stated, "delivering top-tier performance while operating faster and at a lower cost than competing frontier models."
GPT-5.4 continues OpenAI's work to reduce hallucinations and factual inaccuracies. The company reports the new model is 33% less likely to make errors in individual claims compared to GPT-5.2, with overall responses 18% less likely to contain mistakes.
As part of the launch, OpenAI redesigned how the GPT-5.4 API handles tool calling, introducing a new system named Tool Search. Previously, system prompts had to define all available tools upfront—a process that consumed considerable tokens as tool libraries expanded. The new system lets models retrieve tool definitions on demand, making requests faster and more cost-effective in environments with many tools.
OpenAI has also added a new safety evaluation to assess its models' chain-of-thought—the running commentary that reveals the model's reasoning during multi-step tasks. AI safety researchers have long expressed concern that reasoning models might misrepresent their chain-of-thought, and testing confirms this can occur under certain conditions.
OpenAI's new evaluation indicates that deception is less probable in the Thinking version of GPT-5.4, "suggesting the model lacks the capability to conceal its reasoning and that CoT monitoring remains an effective safety tool."
Sam Altman Sparks Debate Over AI's Deceleration
Listen onApple PodcastsListen onSpotifyOpenAI CEO Sam Altman recently suggested that it may be time to “pace the rate of AI development” to allow society to “harden around some of these new capability levels.”On the latest episode of TechCrunch’s Equ
OpenAI fights Apple trade secret lawsuit
OpenAI rebutted Apple’s trade secret allegations on Tuesday, arguing the lawsuit is unfounded.“We take these claims seriously but see no evidence supporting them,” OpenAI stated, as reported by Bloomberg’s Ed Ludlow on X. “We support fair competition
OpenAI robotics head Caitlin Kalinowski resigns over Pentagon partnership
OpenAI robotics leader Caitlin Kalinowski has stepped down following the company’s controversial partnership with the Department of Defense.“This wasn’t an easy call,” Kalinowski explained in a social media statement. “While AI plays a vital role in





Home






