Giving too long prompts won't yield good results from ChatGPT, says OpenAI's new guide

 Giving too long prompts won't yield good results from ChatGPT, says OpenAI's new guide 

OpenAI's new guide disproves the common belief that the longer and more detailed instructions (prompts) given to artificial intelligence (AI) models, the better the results. With the release of the latest version of Chat GPT, GPT-5.6 (GPT-5.6), the company has released a new guide on best prompting practices. According to this, OpenAI itself has confirmed the fact that only clear and simple prompts will give the best results.

According to OpenAI's internal tests, removing unnecessary instructions and using simple prompts improves the evaluation scores of AI models by 10 to 15 percent. In addition, the company claims that total token consumption will decrease by 41 to 66 percent and user costs will decrease by 33 to 67 percent. Research has also shown that limiting Large Language Models (LLMs) to short answers produces more accurate results.

Which model for which job?

OpenAI has introduced various models according to needs under GPT-5.6 series with new naming method. These include the gpt-5.6-sol (flagship model) for superior capacity and heavy-duty work, the gpt-5.6-terra for a balance of affordability and capacity, and the gpt-5.6-luna for high-volume work.

How to make prompting effective?

OpenAI recommends adopting a few key habits to make prompts quick and effective. To find out what is important, you should gradually remove unnecessary things from the prompt and test it. Instead of repeating the same rule in different words, instructions should be clearly given only once. Use tools only where necessary and give examples only when specific reference to the model is required.

Long chats increase the context window and consume a lot of tokens, so keep the chat short. Since the new model is inherently short-answer, saying 'Be concise' in the prompt may not always be effective.

Instead, OpenAI recommends setting the answer length to Low, Medium or High using the text.verbosity parameter in the API. Also, it is better to give clear and concrete instructions to the AI ​​on how to make its tone, rather than giving adjectives like Friendly or Empathetic.

GPT-5.6 can easily handle multiple tasks at once. But the permission and limits to be given to it should be clarified in advance. For example, allowing people to read files and test code on their own, but requiring mandatory permissions for external changes, purchases, or important decisions.

New features of GPT-5.6

The new model features Programmatic Tool Calling for complex and tool-centric tasks, which uses JavaScript to run tools and process results. Similarly, the Multi-agent feature has been introduced in the beta version to run multiple sub-agents at once and complete complex tasks faster. In addition, Explicit prompt caching allows you to choose which prompt to cache. However, the cost of cash writing is 1.25 times more than uncash input.

Pro mode can be used when quality is a priority for difficult tasks. It takes a bit more time and token but gives reliable and same final result with deep analysis. It can also set Max Reasoning Effort, which is useful for very challenging tasks.

Also, through Persisted reasoning, the reasoning of the previous conversation is made reusable in later stages. For visual and photo analysis, the feature of analyzing large photos without resizing has been added under Original image detail.

The latest version of Chat GPT is very good at frontend design, user intent understanding and technical functions. However, to get the best out of these features, OpenAI has made it clear that instead of giving long and vague instructions, it is essential to practice simple, clear and crisp prompting with the right freedom, length and tone.