OpenAI drops GPT-5: smarter, sharper, and built for the real world

More than two years after GPT-4’s release, OpenAI has unveiled GPT-5, boasting sharper reasoning, multimodal input, better math skills, and cleaner task execution, according to the company.

The large language model (LLM) — now rolling out to ChatGPT users and available in the API — is “smarter, more stable, and more versatile” and built to handle real-world tasks more like a human expert, OpenAI said.

[ Related: Open AI: Latest news and insights ]  

In anticipation of OpenAI’s new AI model, Anthropic released the latest version of its own chatbot, Claude, earlier in the week.

Claude Opus 4.1 came with improvements particularly in two key areas: it significantly improved its coding capabilities, solving up to 75% of real-world programming tasks based on SWE Verified benchmarks; and the model is capable in detailed research and analysis, especially in tasks that require tracking lots of information and intelligently finding answers, according to Anthropic.

For developers, OpenAI claims GPT-5 is its most powerful coding model to date, outperforming GPT-o3 in benchmarks and real-world tasks. The model is “fine-tuned for agentic tools” like Cursor, Windsurf, Copilot, and Codex CLI, and it set new records in testing, the company stated in a blog.

According to OpenAI, GPT-5 delivers sharper reasoning, handling complex problems and multi-step instructions with greater accuracy and focus. It stays on track, follows directions more precisely, and produces more useful, reliable output, the company said.

Users will see fewer hallucinations and have better customization tools, making GPT-5 more dependable and easier to adapt to specific industries and needs, OpenAI said.

It also builds on GPT-4o’s multimodal abilities, offering smoother interactions across text, images, and audio, according to OpenAI.

According to research firm Gartner, GPT-5 improves safety and compliance and hasbetter transparency than previous models, making it more enterprise-ready. OpenAI’s newest model offers CIOs a chance to accelerate AI efforts with better reliability and context handling, enabling use cases like document automation and advanced customer interactions. But its impact depends on strong data, governance, and strategy, Gartner said in a report.

Its safe outputs and reasoning aid auditability, though it poses new risks due to stronger reasoning and multimodal abilities, including potential misuse and system bypass. Jailbreak resistance has slightly improved, but API vulnerabilities remain. Gartner warned that human oversight is crucial. “While risk fundamentals haven’t changed, leaders should watch for unsafe integrations and practices like vibe coding that may expose data or bypass guardrails,” it said.

GPT-5 will be OpenAI’s “most significant do or die moment yet,” according to Nathaniel Whittemore, CEO of Superintelligent, a New York-based AI education platform.

“Ever since the launch of ChatGPT, they’ve been the model state of the art. While competitors like Google and Meta can take advantage of hundreds of millions of existing users to put AI products in front of, OpenAI relies on winning new users by being far ahead of the other AI labs,” Whittemore said.

OpenAI chief operating officer Brad Lightcap said ChatGPT is now in use by more than five million business users — up from three million in June.

Biopharmaceutical company Amgen is one of the early adopters of GPT-5. Sean Bruich, senior vice president of AI & Data at Amgen, said AI only works in science if it meets the highest bar, and “GPT-5 clears it,” delivering sharper accuracy, better context, and faster results across Amgen’s workflows.

“GPT-5… is doing a better job navigating ambiguity where context matters. We are seeing promising early results from deploying GPT-5 across workflows,” he said. He also said the model was faster, more reliable, and had higher quality outputs than GPT-4 and other earlier models.

Ethan Mollick, an associate professor at The Wharton School, had early access to GPT-5. “It is a big deal,” he said in a blog post. He asked the model to do something dramatic to prove that point. The model thought for 24 seconds and then delivered a poetic manifesto of AI capability — specifically, a rhetorical, alliterative showcase of “a multifunctional intelligence system.”

“GPT-5 just does stuff, often extraordinary stuff, sometimes weird stuff, sometimes very AI stuff, on its own. And that is what makes it so interesting,” Mollick said.

After “many AI conversations,” Mollick said he has found two big issues that limit most people’s success in using AI models: First, most people don’t know which model to use — so they get fast, weak results instead of more complete answers from the powerful reasoning models.

“The longer [the models] think, the better the answer, but thinking costs money and takes time. So OpenAI previously made the default ChatGPT use fast, dumb models, hiding the good stuff from most users,” Mollick said. “A surprising number of people have never seen what AI can actually do because they’re stuck on GPT-4o, and don’t know which of the confusingly named models are better.”

Second, most people also don’t know what AI can do or how to ask — especially with newer agentic AIs. GPT-5 fixes both problems by choosing models well and suggesting actions, he said. “It is very proactive, always suggesting things to do.”

GPT-5 is beginning to roll out to ChatGPT Plus, Pro, Team, and Free users, with access for Enterprise and Edu customers coming next week. “Once free users reach their GPT‑5 usage limits, they will transition to GPT‑5 mini,” OpenAI said.

More OpenAI news:

• OpenAI challenges rivals with Apache-licensed GPT-OSS models
• Google snatches Windsurf execs in a $2.4B deal, derailing OpenAI’s biggest acquisition yet
• OpenAI and Perplexity enter browser wars to take on Chrome
• OpenAI tests Google TPUs amid rising inference cost concerns
• Microsoft/OpenAI AGI argument unlikely to impact enterprise IT