OpenAI has released GPT-5, its most advanced artificial intelligence model so far, introducing improvements in reasoning, context handling, and task automation. The model is now available to all ChatGPT users in different variants and access tiers.
Model Variants and Routing
GPT-5 is offered in four versions: nano, mini, standard, and chat. An internal routing system selects the most suitable variant for each request, based on its complexity. For more demanding tasks, the GPT-5 thinking mode can be activated to extend reasoning depth.
Performance
OpenAI reports that GPT-5 shows performance comparable to PhD-level knowledge in over 40 professional fields, including law, logistics, sales, engineering, healthcare, and software development. The company says hallucination rates are about 45% lower than in GPT-4o, meaning GPT 5 should make up much less false information than before.
Coding and Benchmark Results
In coding benchmarks, GPT-5 achieved higher scores than previous models:
- SWE-Bench Verified: 74.9%
- Aider Polyglot: 88%
These results surpass GPT-4’s 52% and the O3 model’s 69.1%. The model can chain tool calls and follow longer multi-step instructions, which improves its ability to handle complex workflows such as debugging and generating user interfaces.
Context Window
GPT-5 supports up to 400,000 input tokens for text and vision and produces outputs of up to 128,000 tokens. This capacity enables extended conversations and the analysis of large documents or images in a single session.
Users can adjust GPT-5’s responses by selecting predefined interaction styles: Cynic, Robot, Listener, and Nerd. It also integrates with Gmail and Google Calendar, allowing for direct interaction with emails and schedules within the ChatGPT interface.
Access and Availability
GPT-5 is available immediately:
- Free tier users can access lighter versions with daily limits.
- Plus and Pro subscribers have higher limits and access to the full reasoning capabilities.
- Enterprise, Team, and Education plans are receiving the rollout in stages.
Stay Connected with ProPakistani
Get the latest tech news, telecom insights, and product launches wherever you prefer.
Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.
Oh look more tech hype that’s not even real.
Listen here afaq. Chatgpt has been proven to be wrong multiple times. I uploaded a Cambridge 1996 graduation image showing John Oliver, David mitchell and Richard Ayoade all standing together. The ai didn’t recognize anyone of them. I showed an even clearer image and it still didn’t recognize them.
You don’t need PhD to recognize someone. Ai having PhD knowledge is completely wrong. It’s just fed data from PhD papers. Anyone can do that. Scribed had that over 10 years ago.
Next time use independent benchmark instead of open ai claims.
Afaq being wrong as usual lol
Your other article contradicts you afaq 😆
You lost last bit of your credibility. Even if you had any.