In short
- OpenAI has launched new inner assessments for project-level efficiency.
- Scientific and mathematical benchmarks confirmed greater scores than prior fashions.
- The announcement comes as OpenAI makes offers to combine GPT within the U.S. Authorities and Firms.
Simply weeks after its final main launch, OpenAI is aggressively pivoting its flagship ChatGPT from a client novelty to an indispensable company powerhouse.
On Thursday, the corporate launched GPT-5.2, a brand new giant language mannequin it claims is quicker, extra dependable, and designed to deal with complicated skilled workflows.
The replace sign OpenAI is shifting past homework assist and normal queries, aiming as a substitute to embed its know-how as an important, day by day device within the enterprise world, as evidenced by its profitable offers with the U.S. authorities and Disney.
“We designed GPT‑5.2 to unlock much more financial worth for individuals,” OpenAI mentioned in an announcement. “It’s higher at creating spreadsheets, constructing shows, writing code, perceiving photographs, understanding lengthy contexts, utilizing instruments, and dealing with complicated, multi-step tasks.”
The brand new benchmark for office automation
Touting the efficiency of GPT-5.2, the corporate launched a proprietary analysis benchmark, GDPval, that simulates duties throughout 44 occupations.
GPT-5.2 matched or exceeded human employee efficiency in roughly 71% of the comparisons, the corporate claims.
“On GDPval, the considering mannequin beats or ties human consultants on 70.9% of frequent skilled duties like spreadsheets, shows, and doc creation,” OpenAI CEO of Purposes, Fidji Simo wrote on X. “It’s additionally higher at normal intelligence, writing code, device calling, imaginative and prescient, and long-context understanding so it might unlock much more financial worth for individuals.”
It’s unclear whether or not the benchmark has undergone exterior evaluate, leaving business consultants to attend for impartial verification of the claims.
Technical breakdown: Three fashions for 3 jobs
GPT-5.2 grew to become obtainable throughout paid subscription tiers on Thursday, with API entry opening the identical day. Builders can now select from three distinct variations, every optimized for various skilled wants.
- Immediate: For fast, easy skilled duties.
- Pondering: For extra complicated, multi-step duties.
- Professional: The highest-tier mannequin, constructed for intensive analysis and long-form tasks.
API pricing has been set at $1.75 per million enter tokens and $14 per million output tokens.
Along with the GDPval benchmark, GPT-5.2 confirmed improved efficiency on established technical assessments, posting greater scores on GPQA Diamond and FrontierMath. It additionally reportedly demonstrated extra dependable leads to demanding duties like coding, knowledge evaluation, and experimental design.
Within the announcement, the corporate introduced a number of glowing suggestions statements from early testers.
The discharge of a extra competent office AI arrives in an already tense labor setting.
Company executives seem largely optimistic, with a latest Simply Capital survey exhibiting 93% of enterprise leaders view AI as a optimistic pressure. But, the identical research discovered practically half of Individuals anticipate the know-how to remove jobs, a priority executives reportedly share much less.
Usually Clever E-newsletter
A weekly AI journey narrated by Gen, a generative AI mannequin.