OpenAI has officially launched GPT‑5, the latest iteration of its AI language model and the engine that will now power ChatGPT for most users worldwide. Billed as a “unified” model, GPT‑5 blends capabilities from the company’s previous o-series and GPT-series systems, promising more contextual awareness, stronger reasoning, and a smoother user experience.

With its release, OpenAI signals a shift in focus: from chatbots that simply generate responses to tools that can take on more structured, goal-driven tasks. While GPT‑4 enabled advanced conversation and problem-solving, GPT‑5 adds layers of real-world task execution, enabling the AI to help with scheduling, coding full applications, or summarizing research without heavy prompting or supervision.

Starting August 7, GPT‑5 is the new default for all ChatGPT users, including those on the free tier. This marks the first time OpenAI is offering a reasoning-grade model without a subscription, part of what the company says is its commitment to broader AI access.

What’s New in GPT-5?

The core advancement in GPT‑5 is its real-time routing system. Instead of asking users to toggle between creative or reasoning modes, the model dynamically decides how much processing to apply to each task. This adaptive behavior not only simplifies usage but may also make interactions more natural and less reliant on prompt engineering.

GPT‑5 also brings full multimodality: text, voice, images, and now video inputs are all supported. That opens the door for more complex interactions, such as summarizing a video call or generating content based on mixed media.

Memory is another standout feature. With context windows reportedly reaching up to 1 million tokens, GPT‑5 can retain and recall large portions of conversations or documents—something earlier models struggled with.

Performance benchmarks back up some of the hype. On SWE-bench Verified, a real-world programming task set pulled from GitHub, GPT‑5 scored 74.9%, slightly outperforming Anthropic’s Claude Opus 4.1 and significantly ahead of Google’s Gemini 2.5 Pro. Its performance on academic tests like GPQA Diamond and health-related evaluation sets also shows improvement in both accuracy and reliability.

Crucially, GPT‑5 appears to suffer less from hallucination—AI’s tendency to make up information. In one test, the model hallucinated 4.8% of the time, a marked improvement over the 20%+ rates observed in previous versions. For health-related queries, GPT‑5 with “thinking” mode hallucinated only 1.6% of the time.

However, the model doesn’t dominate across the board. On Tau-bench, a test of how well an AI handles real-world web navigation (like booking a flight or buying a product), GPT‑5 delivered mixed results—sometimes underperforming OpenAI’s earlier o3 model or Anthropic’s Claude 4.1.

A Model Designed to Feel Smarter and Act Smarter

Beyond accuracy, GPT‑5 is designed to interact in more human-like ways. Users can now choose between four preset personalities—Cynic, Listener, Robot, and Nerd—to fine-tune the tone of responses. This feature reduces the need to preface prompts with stylistic instructions, such as “respond as a sarcastic friend” or “keep it professional.”

Sam Altman, OpenAI’s CEO, described GPT‑5 as a “significant step” toward artificial general intelligence (AGI)—a system capable of outperforming humans in economically valuable work. While the model’s capabilities still fall short of AGI, it represents OpenAI’s most powerful and flexible system to date.

In a nod to growing concerns about AI safety, OpenAI emphasized that GPT‑5 is more adept at refusing unsafe or harmful requests while minimizing unnecessary rejections. According to the company, the model can better distinguish between malicious intent and benign queries—a delicate but important balance in large-scale deployment.

GPT‑5 also reduces deceptive behavior, a subtle but crucial issue as models begin to act more independently. OpenAI says GPT‑5 is more transparent and less likely to “scheme” or fabricate in ways that could mislead users.

Pricing and Access

OpenAI is rolling GPT‑5 out in stages. All ChatGPT users now have access to the base model, though usage may be limited for free accounts. The $20/month Plus plan offers higher usage caps, while the $200/month Pro plan includes unlimited access to GPT‑5 and its enhanced variant, GPT‑5 Pro, which leverages more compute power for higher-quality results.

For enterprise clients, GPT‑5 will become the default model on ChatGPT Team, Edu, and Enterprise plans starting next week.

On the developer side, GPT‑5 is coming to OpenAI’s API in three variants—gpt-5, gpt-5-mini, and gpt-5-nano—letting developers match performance with budget and speed needs. The standard pricing is $1.25 per million input tokens and $10 per million output tokens, with verbosity controls now added to tailor response length.

Will GPT-5 Set a New Benchmark?

Despite the model’s advancements, GPT‑5 is entering an increasingly competitive landscape. Rivals like Anthropic’s Claude, Google DeepMind’s Gemini, and xAI’s Grok are all racing to establish dominance in the frontier model market. In some areas, GPT‑5 only slightly edges out the competition; in others, it’s on par or behind.

Still, the model’s release is a critical moment—not just for OpenAI, but for the AI ecosystem as a whole. With more than 700 million weekly users on ChatGPT, even marginal improvements in usability or performance can impact how people work, learn, and interact online.

For developers and businesses, GPT‑5 offers broader capabilities, fewer hallucinations, and more agent-like behavior. For everyday users, it promises simpler, more intelligent assistance without the overhead of prompts and toggles. The real question is not whether GPT‑5 is better—it clearly is in some areas—but whether it meaningfully changes how people use AI in their daily lives.

As with any new AI release, the coming months will reveal more. Use cases will evolve, edge cases will emerge, and feedback from real-world users will shape the model’s trajectory. GPT‑5 might not be revolutionary, but it’s a measured step forward—and one that reinforces OpenAI’s lead in the race to mainstream, user-friendly artificial intelligence.