Here Are All The Gpt-5 Updates Openai Has Rolled Out Since Launch

SOPA Images/Contributor/Getty

ZDNET's cardinal takeaways:

• OpenAI released its long-awaited GPT-5 connected Thursday.
• Some users complained GPT-5 was inferior to its predecessor, 4o.
• In response, nan institution announced a flurry of changes.

OpenAI released GPT-5, nan long-awaited upgrade to nan exemplary which powers ChatGPT, Thursday. In emblematic OpenAI fashion, nan merchandise has included plentifulness of twists, turns, and drama.

It was almost inevitable that nan caller exemplary would disappoint a important number of people, fixed really intensely it was hyped starring up to its release. On Aug. 2, Xikun Zhang, a investigation intelligence astatine OpenAI, wrote successful an X station that "GPT-5 really feels for illustration nan astir anticipated merchandise motorboat successful history." (The station was later deleted.)

(Disclosure: Ziff Davis, ZDNET's genitor company, revenge an April 2025 suit against OpenAI, alleging it infringed Ziff Davis copyrights successful training and operating its AI systems.)

Also: How you tin still entree GPT-4o, o3, and different older models successful ChatGPT

Sure enough, GPT-5 has debuted much pinch a whimper than a bang.

GPT-5's halfway strength, which OpenAI has gone to awesome lengths to emphasize, is its expertise to adaptively take betwixt a postulation of models depending connected nan quality of a peculiar prompt. While a bully thought successful theory, it's turned retired to beryllium unwieldy successful believe for users, galore of whom are now requesting retired of vexation that nan institution bring backmost 4o, GPT-5's predecessor model. GPT-5 also performed poorly successful coding tests conducted by ZDNET.

It didn't thief that aggregate charts displayed inaccurate accusation during a livestream position of GPT-5 connected Thursday, a slip-up that nan net has taken to deriding arsenic nan "chart crime." One barroom chart comparing nan caller model's capacity connected nan SWE-bench test, for illustration -- which gauges models' coding skills -- showed a comparatively debased people for GPT-5 being displayed wrong a disproportionately higher bar.

Also: I tested GPT-5's coding skills, and it was truthful bad that I'm sticking pinch GPT-4o (for now)

All of this has presumably led to a engaged fistful of days for nan OpenAI engineering and media teams arsenic they woody pinch nan fallout of this latest snafu. A postulation of X posts some from nan institution and its CEO, Sam Altman, since Thursday, provides a model into really nan institution has responded, and nan various patches that person been applied to support nan leaky vessel of GPT-5 afloat.

What has been updated?

A Friday X post from Altman included a drawstring of updates that had been made to GPT-5 successful consequence to early personification feedback. Those included:

• Doubling nan complaint limit for GPT-5 for ChatGPT Plus subscribers. According to OpenAI's website, Plus users tin now nonstop up to 160 messages each 3 hours via GPT-5. Once that limit has been hit, nan exemplary will automatically move to a much constricted mini exemplary until nan usage clip model resets. OpenAI besides noted that this was "a impermanent summation and will revert to nan erstwhile limit successful nan adjacent future."

• Allowing immoderate paid subscribers to proceed to entree and usage 4o. In a X post connected Friday, OpenAI wrote that ChatGPT Plus and Team users tin entree nan exemplary done an action reference "show bequest models" successful settings. The aforesaid station added that GPT-5 had been successfully rolled retired to each ChatGPT Plus, Pro, Team, and free users.

• Making it easier for users to spot which exemplary is being utilized to respond to a fixed query. In nan aforesaid Friday X post, OpenAI said users tin now find this accusation by hovering complete nan information arrow icon connected nan acold correct of nan paper appearing beneath each of nan chatbot's responses.

• Adding measures to guarantee nan correct exemplary is utilized for each response. (Altman noted successful his Friday station that nan model's autoswitcher system was surgery for portion of nan erstwhile day, truthful that it "seemed measurement dumber.")

Next steps

In another X post connected Friday, Altman said nan early, less-than-stellar reception to GPT-5 had imparted important lessons astir personification interface and acquisition that would beryllium taken into relationship arsenic ChatGPT continued to evolve.

Also: OpenAI's GPT-5 is now free for all: How to entree and everything other we know

"Long-term, this has reinforced that we really request bully ways for different users to customize things," he wrote. "For a silly example, immoderate users really, really for illustration emojis, and immoderate ne'er want to spot one. Some users really want acold logic and immoderate want warmth and a different benignant of affectional intelligence."

Altman went connected to opportunity that nan institution was "doing heroic work" to build an optimal personification acquisition amid nan perpetually shifting demands of a awesome merchandise rollout, each nan while moving wrong nan constraints afforded by its proviso of GPUs: "Not everyone will for illustration immoderate tradeoffs we extremity up with, obviously, but astatine slightest we will explicate really we are making decisions."