OpenAI’s weblog submit claims that GPT-5 beats its earlier fashions on a number of coding benchmarks, together with SWE-Bench Verified (scoring 74.9 %), SWE-Lancer (GPT-5-thinking scored 55 %), and Aider Polyglot (scored 88 %), which check the mannequin’s capacity to repair bugs, full freelance-style coding duties, and work throughout a number of programming languages.
In the course of the press briefing on Wednesday, OpenAI post-training lead Yann Dubois prompted GPT-5 to “create an exquisite, extremely interactive net app for my companion, an English speaker, to study French.” He tasked the AI to incorporate options like each day progress, quite a lot of actions like flashcards and quizzes, and famous that he wished the app wrapped up in a “extremely participating theme.” After a minute or so, the AI-generated app popped up. Whereas it was only one on-rails demo, the outcome was a glossy website that delivered precisely what Dubois requested for.
“It is an awesome coding collaborator, and in addition excels at agentic duties,” Michelle Pokrass, a post-training lead, says. “It executes lengthy chains and power calls successfully [which means it better understands when and how to use functions like web browsers or external APIs], follows detailed directions, and offers upfront explanations of its actions.”
OpenAI additionally says in its weblog submit that GPT-5 is “our greatest mannequin but for health-related questions.” In three OpenAI health-related LLM benchmarks—HealthBench, HealthBench Laborious, and HealthBench Consensus—the system card (a doc that describes the product’s technical capabilities and different analysis findings) states that GPT-5-thinking outperforms earlier fashions “by a considerable margin.” The considering model of GPT-5 scored 25.5 % on HealthBench Laborious, up from o3’s 31.6 % rating. These scores are validated by two or extra physicians, in response to the system card.
The mannequin additionally allegedly hallucinates much less, in response to Pokrass, a typical challenge for AI the place it offers false data. OpenAI’s security analysis lead Alex Beutel provides that they’ve “considerably decreased the charges of deception in GPT-5.”
“We’ve taken steps to scale back GPT-5-thinking’s propensity to deceive, cheat, or hack issues, although our mitigations usually are not good and extra analysis is required,” the system card says. “Particularly, we’ve educated the mannequin to fail gracefully when posed with duties that it can’t clear up.”
The corporate’s system card says that after testing GPT-5 fashions with out entry to net searching, researchers discovered its hallucination charge (which they outlined as “proportion of factual claims that comprise minor or main errors”) 26 % much less frequent than the GPT-4o mannequin. GPT-5-thinking has a 65 % diminished hallucination charge in comparison with o3.
For prompts that might be dual-use (probably dangerous or benign), Beutel says GPT-5 makes use of “secure completions,” which prompts the mannequin to “give as useful a solution as potential, however inside the constraints of remaining secure.” OpenAI did over 5,000 hours of pink teaming, in response to Beutel, and testing with exterior organizations to ensure the system was strong.
OpenAI says it now boasts practically 700 million weekly lively customers of ChatGPT, 5 million paying enterprise customers, and 4 million builders using the API.
“The vibes of this mannequin are actually good, and I feel that persons are actually going to really feel that,” head of ChatGPT Nick Turley says. “Particularly common individuals who have not been spending their time eager about fashions.”
