OpenAI launches GPT-5: What you need to know about the new model and its controversial debut

On August 7, 2025, OpenAI officially unveiled GPT-5 — its newest and most advanced language model, which the developers describe as a significant step toward artificial general intelligence (AGI). The model has been made available to all ChatGPT users, including free accounts as well as Plus, Pro, and Team subscribers.

However, the launch was met with mixed — and largely negative — reactions, including on Reddit, alongside criticism over errors in the presentation and charts that raised questions within the community. Here’s what GPT-5 is, what it can do, and why its debut feels so divisive.

What is GPT-5?
OpenAI presents GPT-5 as “the smartest, fastest, and most useful” model in its history. The company’s CEO, Sam Altman, called it “a significant step toward AGI,” though he clarified that the model has not yet reached the level of artificial general intelligence as defined in OpenAI’s charter — “a highly autonomous system that outperforms humans at most economically valuable work.” Among its limitations, Altman noted the absence of the ability for continuous learning after deployment.

GPT-5 is not a single model but an integrated system with multiple variants that are automatically selected depending on the complexity of the request and the user’s subscription type. The model family includes:

• gpt-5-main: Standard fast model for most tasks.
• gpt-5-main-mini: A simplified, faster version.
• gpt-5-main-nano: A compact, low-cost model available only via API.
• gpt-5-thinking: A “reasoning” model for complex tasks such as programming and analytics.
• gpt-5-thinking-pro: An accelerated deep-analysis version for Pro subscribers.

A built-in router analyzes the query and directs it to the appropriate model, streamlining interaction and improving efficiency. GPT-5 features a 256,000-token context window (compared to 200,000 in the previous o3 model), allowing it to better process long texts, code, or dialogues without losing context.

Key Features
OpenAI claims GPT-5 outperforms its predecessors in accuracy, speed, and usefulness, with fewer “hallucinations” — instances where the model outputs false information. Key improvements include:

• Programming: GPT-5 achieved high scores on benchmarks — 74.9% on SWE-Bench Verified, 55% on SWE-Lancer (in thinking mode), and 88% on Aider Polyglot. It can create complex applications with minimal prompting, fix bugs, and work in multiple programming languages. In a press conference demo, GPT-5 built an interactive French-learning web app with flashcards, quizzes, and progress tracking in under a minute.
• Medicine: GPT-5 improved results on HealthBench, HealthBench Hard, and HealthBench Consensus benchmarks, scoring 25.5% on HealthBench Hard (vs. 31.6% for o3). OpenAI emphasizes that the model does not replace a doctor but can help interpret medical data and prepare questions for specialists.
• Personalization: Users can choose from four personality presets — Cynic, Robot, Listener, or Nerd — and customize chat colors. This feature will soon be available in voice mode.
• Integrations: Starting next week, Pro subscribers will be able to connect Gmail, Google Contacts, and Google Calendar so ChatGPT can automatically use data from these services.

OpenAI also reported that the model underwent over 5,000 hours of safety testing, including “red teaming” with external organizations. For potentially dangerous requests, GPT-5 uses “safe completions” to provide the most useful answer within safe boundaries.

Availability and Limits
GPT-5 is available to all ChatGPT users but with tier-based limits:

• Free account: 10 queries to the basic GPT-5 every 5 hours, after which it switches to gpt-5-mini. One query per day to GPT-5-thinking.
• Plus: 80 queries to GPT-5 every 3 hours, up to 200 thinking-mode queries per week. The model auto-switches to reasoning mode when needed.
• Pro and Team: Unlimited access to all versions, including gpt-5-thinking-pro, as long as there’s no abuse (e.g., mass data processing).
• Enterprise and Edu: Access will roll out within a week of launch.

For developers, GPT-5 is available via API at $1.25 per 1M input tokens and $10 per 1M output tokens. Mini and nano models are significantly cheaper: $0.25/$2 and $0.05/$0.40 per 1M tokens respectively.

Community Reaction: Mixed Reviews
Despite OpenAI’s bold statements, GPT-5’s launch received a lukewarm — often negative — reception on Reddit. Users expressed disappointment for several reasons:

• Lack of “wow factor”: Many felt GPT-5 was only an incremental update over GPT-4o, GPT-4.5, and o3.
• Restrictions on model choice: OpenAI’s plans to remove manual model selection (e.g., o3) drew criticism. Users believe the automatic router may save the company resources but reduces flexibility.
• Access issues: Some users, including Plus subscribers, reported being unable to access GPT-5 despite OpenAI’s promise of immediate availability.

The launch was also marred by errors in charts shown during the presentation, which sparked ridicule and accusations of “chart crimes.” For example, in one chart displaying SWE-Bench Verified results:

• The bar for GPT-5 without thinking mode (52.8%) was taller than for o3 (69.1%) despite having a lower score.
• The bars for GPT-4o (30.8%) and o3 (69.1%) were the same height despite the large difference in values.

These mistakes fueled criticism for misleading the audience. Sam Altman acknowledged the issue, calling it a “mega chart screw-up,” and an OpenAI marketing staffer apologized, stating the correct version had been posted on the company blog. Nevertheless, Reddit and other communities viewed this as a serious misstep for a company that prides itself on accuracy and scientific rigor.

Bottom Line
GPT-5 is a powerful step forward for OpenAI, offering improvements in programming, medical query handling, and personalization. The model promises greater accuracy and utility thanks to reduced hallucination rates and an expanded context window. However, the launch has been overshadowed by technical presentation flaws and mixed community feedback, with many expecting more revolutionary changes. Chart errors and limited access for some users have deepened skepticism — especially amid strong competition from Anthropic, Google, and others.

In the coming weeks, it will become clear whether GPT-5 meets expectations in real-world scenarios such as business automation, scientific research, and app development. For now, OpenAI faces the challenge of regaining community trust by fixing flaws and proving that GPT-5’s “PhD-level intelligence” is more than just a marketing slogan — but a genuine value for users.