GPT-5 explained: Inside OpenAI's most advanced AI model yet

Enhanced Writing and Coding

GPT-5 is not just a better writer—it’s a better builder. “The most significant improvement appears to be in real-world coding tasks, where GPT5 seems to be at least as good or better than Anthropic's Claude 4 models—Sonnet 4 and Opus 4/4.1—which have been the leaders so far. It is also significantly cheaper than competing models,” says Parav Nagarsheth, founder of AI companion startup Fondant. He adds, “Like Sam Altman rightly said, they seem to have pushed for real-world use-cases and affordability.”

Nagarsheth admits he is yet to take a closer look at GPT-5, but says, “looking at early takes, it seems it is really good at one-shotting prototypes of complex web demos. It also appears to be good at navigating complex codebases, at least at par with Anthropic's models.”

Higher Reliability

Accuracy has long been a challenge for AI models, but GPT-5 significantly raises the bar. It reduces factual errors—often referred to as “hallucinations”—by 45–80 percent compared to GPT-4o. "There's always a need for a human-in-the-loop with such models especially when dealing with conversations in highly regulated industries like fintech," says Anshul Mittal, lead engineer, Prodigal, an AI Agent startup for loan servicing and collections. "In the demo they mentioned that they created new evaluation metrics to track hallucination and showed how GPT-5 scores best in that. We will have to try this out and see how these benchmark tests translate to real world use-cases," he adds.

Deeper Reasoning

GPT-5 introduces a more thoughtful approach to problem-solving by using internal chains of reasoning. Instead of jumping straight to an answer, it breaks down complex queries into logical steps—much like how a human expert would think through a problem. This makes it especially effective for tasks that require multi-step analysis, across sectors.

Custom Instructions

OpenAI is rolling out a research preview of four pre-set personalities for ChatGPT, giving users more control over how the AI communicates. “The four initial options, Cynic, Robot, Listener, and Nerd, are opt-in, adjustable anytime in your Customize ChatGPT settings, and designed to match your communication style,” states the press release.

Improved Synthesis

Lastly, GPT-5 excels at integrating information from multiple sources—uploaded files, connected apps like Gmail or Google Drive, and even previous chat history. This means it can read and analyse a spreadsheet, cross-reference it with a PDF report, and summarise key insights in one go. It is particularly useful for research, reporting, and decision-making, where pulling together data from different formats and contexts is essential.

Comparison with DeepSeek

GPT-5, OpenAI’s flagship model, represents a scaling-first approach: Massive compute, multimodal capabilities, and deep reasoning. It’s designed for high-performance tasks across industries, from coding and writing to autonomous agents.

In contrast, DeepSeek—especially with its DeepSeek-R1 and DeepSeek-V3 models—focuses on efficiency and openness. These models are optimised for reasoning and cost-effectiveness, and are released under open-source licenses, making them attractive to developers and organisations seeking transparency and affordability. "DeepSeek is yet to see a wider adoption because of the challenges it has faced from a privacy and security standpoint. Despite companies offering self-hosted versions of these models, there's still some pushback in organisations from using the model," adds Mittal.

Launched earlier this week, OpenAI’s GPT-OSS models (like GPT-OSS-120B and GPT-OSS-20b) are more direct competitors to DeepSeek’s open-source offerings. OpenAI is setting its sights on AI builders and developers seeking to rapidly build on real-world use cases on their own systems with its GPT-OSS models. GPT-5, meanwhile, competes with DeepSeek-V3 and R1 in terms of capability and adoption, especially in enterprise and developer ecosystems.