Earlier last week, we reported on the upcoming release of GPT-5. Now, the launch is official and the capabilities are even more impressive than expected. OpenAI’s smartest AI can now build complete applications from simple prompts, deliver expert-level answers across complex topics, and, for the first time, it’s available to the public for free.
Sam Altman, OpenAI’s CEO, tried switching back to GPT-4 after using GPT-5. His verdict? “It was quite miserable.” When the person who built both models can’t go backwards, you know something fundamental has shifted.
Advanced AI reasoning previously cost $200 monthly through ChatGPT Pro. Now everyone gets access to the same technology that was exclusively available to premium subscribers just days ago. Free users finally have their hands on the AI that thinks before it responds, builds instead of just suggesting, and knows when to work harder on complex problems.
GPT-5 Builds Apps Faster Than Developers
During the launch demonstration, OpenAI showed GPT-5’s most impressive trick which is “vibe coding.” Researchers asked the AI to create a French learning app for English speakers. The requirements were complex: flashcards, quizzes, progress tracking, and an engaging theme. GPT-5 delivered two completely different, fully functional web applications within seconds.
The model scored 74.9% on SWE-bench Verified, a tough test using real GitHub coding challenges. This puts GPT-5 slightly ahead of Anthropic’s Claude Opus 4.1 (74.5%) and significantly better than Google’s Gemini 2.5 Pro (59.6%).
Box CEO Aaron Levie tested GPT-5 extensively before launch. Previous AI models failed his company’s hardest tests because they couldn’t handle complex math or logic in long documents. “GPT-5 is a complete breakthrough,” Levie told CNBC. “The model retains way more information and uses much higher reasoning capabilities to make decisions.”
Free Users Get Premium Features Without Paying
OpenAI’s decision to give free users access to GPT-5 breaks their previous business model. Free users previously only accessed basic models while advanced reasoning was locked behind paywalls. Now everyone gets the company’s smartest AI, though with usage limits.
The pricing structure works like this:
- Free users: Access to GPT-5 and GPT-5 mini with limited usage
- Plus subscribers ($20/month): Higher usage limits
- Pro subscribers ($200/month): Unlimited access plus GPT-5 Pro for the hardest problems.
When free users hit their limits, they automatically switch to GPT-5 mini, a smaller but still capable version. This ensures continuous access without cutting users off completely.
Medical Questions Get PhD-Level Answers
The model scored significantly higher than previous versions on HealthBench, evaluations based on real medical scenarios and physician-defined criteria. When thinking through complex health questions, GPT-5 hallucinates (provides false information) only 1.6% of the time compared to 12.9% for GPT-4o and 15.8% for the previous o3 model.
The model acts as an active thought partner, flagging potential concerns and asking questions to provide more helpful responses. It adapts answers based on users’ context, knowledge level, and geographic location. OpenAI emphasizes that ChatGPT doesn’t replace medical professionals but helps users understand results and prepare better questions for doctors.
Smart Router Eliminates Model Selection Confusion
GPT-5 introduces a unified system that automatically decides how to handle each request. Users no longer need to choose between different models or settings. The system includes:
- A fast, efficient model for quick responses
- A deeper reasoning model (GPT-5 thinking) for complex problems
- A real-time router that picks the right approach based on conversation complexity
The router continuously learns from user behavior, including when people switch models manually, response preferences, and measured accuracy. This creates a smoother experience where the AI automatically applies the right level of intelligence to each task.
Hallucinations Drop by 80% with New Safety Approach
OpenAI tackled one of AI’s biggest problems which is making up information. GPT-5 with thinking shows 80% fewer hallucinations than the o3 model and 45% fewer factual errors than GPT-4o when using web search.
The company moved beyond simple refusal training to “safe completions.” Instead of just saying no to potentially risky questions, GPT-5 provides helpful high-level information while staying within safety boundaries. For dual-use topics like virology, the model can safely answer general questions but won’t provide dangerous details.
GPT-5 also became more honest about its limitations. When researchers removed images from visual reasoning tests, the previous o3 model still gave confident answers about non-existent images 86.7% of the time. GPT-5 correctly identified missing information 91% of the time.
Microsoft Integration Brings GPT-5 Everywhere
Microsoft moved quickly to integrate GPT-5 across its ecosystem. The new model powers Microsoft 365 Copilot, consumer Copilot, and Azure AI Foundry for developers. CEO Satya Nadella noted the rapid progress since GPT-4’s debut in Bing just two and a half years ago.
GitHub Copilot also gained GPT-5 access for all paid plans, giving millions of developers access to the improved coding capabilities immediately.
The Road to 700 Million Users
ChatGPT reaches nearly 700 million weekly active users, with 5 million business customers and 4 million developers using OpenAI’s APIs. The company expects to hit these numbers this week as GPT-5 rolls out globally.
OpenAI is reportedly in discussions with investors about a potential stock sale that could value the company at around $500 billion, putting it among the most valuable private companies in the world.
OpenAI is proving that Artificial intelligence will actively help people more smartly, rather than replying to prompts only. OpenAI is betting that mass access will speed up adoption and firmly secure ChatGPT’s place as the leader in the AI space.
Comments 2