The GPT-5 Cheat Sheet: 13 Issues to Know About OpenAI's Newest AI Leap – Decrypt

OpenAI unveiled GPT-5 throughout a Thursday livestream, marking what the corporate known as a qualitative shift in synthetic intelligence functionality after a number of months of anticipation and a number of delays. The mannequin is being rolled out to all ChatGPT customers all through the day right now.

The discharge represents OpenAI’s try to unify its varied AI applied sciences right into a single system. The corporate described reasoning as central to its synthetic normal intelligence technique, with the breakthrough eliminating earlier trade-offs between velocity and analytical depth. Customers not want to decide on between quick responses and deep reasoning capabilities—GPT-5 delivers each concurrently.

Right here’s a cheatsheet on what you’ll want to know.

1. When can I get it?

GPT-5 rolls out right now on ChatGPT and through its API. Microsoft has additionally included GPT-5 into its merchandise instantly, making it out there via Copilot and GitHub Copilot.

For those who up to date your Edge browser with Copilot, you ought to be prepared to make use of it now.

2. Does everybody get the identical model?

Sure, form of: Free tier customers will begin with the usual GPT-5 earlier than transitioning to a lighter “GPT-5 mini” model once they deplete their utilization quota. Professional subscribers ($200 a month) get limitless entry to the complete mannequin whereas Plus subscribers ($20/month) get entry to straightforward GPT-5.

Professional subscribers can run GPT-5 at its highest intelligence stage with further options like early entry to its superior brokers, limitless utilization, extra capabilities for deep analysis, precedence entry, and superior voice mode with larger limits for video and display screen sharing.

3. What does multimodal imply? Does a separate picture generator go away?

Multimodal means GPT-5 can course of and generate various kinds of content material—textual content, pictures, voice, and now even video—all throughout the similar dialog. The mannequin confirmed enhanced overseas language understanding for advanced duties, producing full web sites with French phrases and correct pronunciation.

As a substitute of juggling between Imaginative and prescient, Sora, GPT, and the “o” fashions to cause, GPT-5 can do every thing by itself.

4. How large is the context window and why does it matter?

GPT-5 has a 256,000 token context window for enter, with the API accepting as much as 272,000 enter tokens and emitting a most of 128,000 reasoning and output tokens, for a complete context size of 400,000 tokens.

This implies it might course of roughly 200,000 phrases without delay—equal to an extended novel. The bigger context window permits GPT-5 to keep up coherent conversations over for much longer interactions and analyze total codebases or prolonged paperwork with out shedding monitor of essential particulars.

That mentioned, this window isn’t very large by right now’s requirements. Only for context, Gemini 2.5 is able to dealing with 1 million tokens,

5. What new options does it have?

None, actually, however a few of its expertise are upgraded to such a level that they may really feel like new options.

6. So what’s so nice about it?

GPT-5 is extra highly effective in nearly each approach. As an example, it demonstrated outstanding coding capabilities in the course of the presentation, writing over 400 traces of code in two minutes when prompted to create a Bernoulli impact simulation from scratch. Different cool issues proven off within the demo:

• Voice interactions sound much less robotic and stay video capabilities have been launched that match opponents like Gemini Reside.
• The mannequin can now analyze uploaded pictures and incorporate them into its responses.
• It’s higher at agentic duties and is supposedly in a position to deal with real-world functions and clarify its reasoning.
• Subsequent week customers will be capable to combine Gmail and Google calendar, which is able to permit it to be a a lot better assistant.

7. Has pricing modified?

ChatGPT subscription pricing stays unchanged at $20/month for Plus and $200/month for Professional.

For API customers, GPT-5 prices $1.25 per million enter tokens and $10.00 per million output tokens for the usual mannequin. GPT-5 mini prices $0.25 per million enter tokens and $2.00 per million output tokens, whereas GPT-5 nano runs $0.05 for enter and $0.40 for output.

This is able to make the mannequin aggressive in opposition to choices from different corporations and even cheaper than different fashions from OpenAI like GPT-4.1 or OpenAI o1 professional which prices a whopping $600 per million tokens.

8. Are we at AGI but?

No. Nevertheless, the corporate positioned reasoning as “on the coronary heart of our AGI program.”

The mannequin represents important progress however stays targeted on particular duties moderately than matching human intelligence throughout all domains. As an example, GPT-5 is nice at language duties however lacks the final intelligence required to carry out a variety of actions independently. It’s not but self-teaching or self-adapting.

9. Can GPT-5 generate movies?

Not but. Whereas video era wasn’t included within the preliminary launch, OpenAI has Sora for video creation as a separate product.

CEO Sam Altman beforehand indicated that future variations would help video “finally.”

The present model does perceive stay video, nevertheless, so it might watch you attempt to repair a motorcycle and supply stay directions.

10. How dependable is it in comparison with earlier fashions?

OpenAI reported that GPT-5 is “considerably much less misleading” than earlier fashions, addressing some of the persistent challenges in massive language mannequin deployment.

On factual accuracy benchmarks, GPT-5 makes roughly 80% fewer factual errors than o3, making it considerably extra reliable for enterprise functions in response to Jakub Pachocki, OpenAI’s chief scientist.

11. What about reminiscence and personalization?

GPT-5 supposedly will provide higher persistent reminiscence throughout classes, remembering information, preferences, and directions throughout a number of conversations, even if you happen to shut the app and open a brand new tab days later. GPT-4’s reminiscence was restricted, particularly days after a session paused.

The corporate mentioned now you can set long-term aims (e.g. assist me lose 10 kilos in a wholesome approach, or assist me put together for my physics check), and GPT-5 will adapt its responses accordingly to proactively align along with your targets.

12. How personal is my private knowledge?

Altman beforehand acknowledged that OpenAI may need handy over a consumer’s private knowledge to the federal government if legally required to take action.

13. Do I want to change between totally different fashions anymore?

Not anymore—except you wish to generate video through Sora. With GPT-5’s launch, OpenAI expressed confidence in deprecating all earlier fashions.

The corporate designed GPT-5 to deal with all use circumstances that beforehand required specialised fashions, although customers can nonetheless select between GPT-5, GPT-5 mini, and GPT-5 nano based mostly on velocity and price necessities.