ChatGPT in one minute
Ask ChatGPT to explain a text, compare documents or prepare a working file. Start with a specific outcome, supply reliable information and identify what needs checking. A fluent answer is not automatically a verified fact. Chat, Work, Codex and the API also provide different environments and access arrangements.
What is ChatGPT?
ChatGPT is an assistant you can talk to while working towards an outcome. You can refine a question, add a file, request changes and ask follow-up questions. Models and product features support this experience. Availability depends on your plan, device, region, workspace and the current rollout.
The distinction from GPT is straightforward: GPT is a model family, while ChatGPT is a product using models. A new GPT version does not mean that ChatGPT has been relaunched or every feature has immediately changed models. An account type also does not fully identify the model handling a particular task. Check the model selection where it is available.
ChatGPT is not an automatic guarantee of accuracy. A clear answer may contain an incorrect fact, reference or calculation. Use checks that match the task. A draft birthday invitation needs a different level of scrutiny from a customer quotation containing many product codes, quantities and delivery conditions.
History: from GPT-3 to today’s ChatGPT
GPT-3 was an earlier language-model milestone, described in OpenAI research published on 28 May 2020. It demonstrated how varied tasks could be supplied through instructions and examples. The later GPT-3.5 series provided the model basis for the first ChatGPT. The model and product histories should therefore be kept distinct.
OpenAI launched ChatGPT on 30 November 2022 as a research preview. The conversational format made follow-up questions and corrections accessible. It was a new product built on earlier model development, rather than a product whose name simply matched the number of a preceding GPT generation.
GPT-4 followed on 14 March 2023. Its announcement described a multimodal model, while image-input access was expanded in stages. GPT-4o arrived on 13 May 2024, marking a significant step in combining text, images and audio. Announcing a model and making every associated feature available were not necessarily simultaneous events.
o1-preview, introduced on 12 September 2024, gave greater prominence to processing before answering. Reasoning models were an important intermediate development, rather than a renaming of ChatGPT itself. GPT-5 was introduced on 7 August 2025. Separately, the introduction of Codex as a cloud agent on 16 May 2025 marked a product development in programming.
The GPT-5.6 family was announced for general availability on 9 July 2026, with Sol, Terra and Luna. The API changelog records GPT-6 Astra on 3 September 2026, GPT-6 Sol and Luna on 22 September, and GPT-6.1 Sol on 29 September. Several generations coexist at the review date. Current user documentation also distinguishes models in Work and Codex from those in Chat. This timeline highlights major changes rather than listing every release.
Writing, explanations and everyday questions
One simple use is rewriting difficult material for a particular reader. Explain who the reader is and what the text should achieve, rather than just asking for better wording. A new customer and an experienced employee need different levels of detail. Identify the facts, terminology and conditions that must remain unchanged.
Suppose a supplier provides a technical description. Ask for a clear version followed by a list of unresolved information. This keeps missing data visible. Do not ask for extra benefits unless you can substantiate them. Persuasive wording can otherwise introduce a product claim that the original information never supported.
ChatGPT can also help with translation, structure and learning. Ask for a summary of a difficult passage, then ask which exceptions a shorter version might lose. A glossary helps preserve product names and approved terms in translation. For learning, request examples and practice questions, then explain what remains unclear. The interaction lets you improve the fit of the explanation.
Working with images and files
Suitable features let you discuss images and generate or edit pictures. A photograph might help describe visible components, for example. Image understanding and image generation are different capabilities. An assistant that analyses a picture does not necessarily use the same model to create a new one; the product can call a separate tool.
For files, define the task precisely. Summarising a PDF, comparing two versions and checking a spreadsheet require different work. Identify the authoritative file and request references to pages, sections or rows where possible. Scans with unclear numbers need particular care. Ask the assistant to flag uncertainty instead of silently filling missing characters.
Consider comparing two supplier catalogues. Start with matching product codes, then compare package sizes, then request an overview. This reduces the risk of treating similar names as identical items. For spreadsheets, specify whether formulas, values and formatting should be preserved. An attractive output is useful only if the underlying information and calculations are correct.
A sample file can help with documents and presentations. Explain what is fixed and what may change. Inspect the final layout as well: correctly copied text can still overflow a table or break awkwardly across pages. File creation does not remove the need for that final check. Delivering a concrete result is a central part of Work.
Web search and deeper research
Answering from model knowledge is different from consulting current sources. For prices, availability and recent events, pay attention to search use and references. Open important sources to check that they support the claim. A plausible link may concern a different period or a narrower question than the response suggests.
A search can establish a specific current fact. A research task goes further by examining several sources, explaining differences and preparing a supported overview. ChatGPT offers research features whose access and usage limits vary by plan. Their names do not guarantee complete or error-free investigation of every topic.
For example, to compare three software packages for a small shop, supply the required features, language and business context. Ask for official sources, a review date and a separate column for unconfirmed points. Distinguish vendor marketing from documented functionality. This produces a decision aid rather than a ranking with invisible criteria.
Scope matters in marketing research too. Studying five pages provides evidence about those pages, not automatically an entire market. Ask which sources were consulted and what remained inaccessible. Good research output makes the basis of a conclusion visible and identifies where further evidence or human checking is required.
Voice and spoken interaction
ChatGPT has voice features, with capabilities and limits depending on plan and platform. Spoken interaction can be useful for exploring ideas, practising a language or asking a question while occupied with another task. Live conversation, recording transcription and reading a file aloud are distinct processes, even when they appear in the same product.
For language practice, agree on a scenario such as a customer conversation, then request focused feedback. Asking for two improvements after a round keeps the exercise manageable. For work notes, check the resulting text for names, commitments and numbers. A misunderstood phrase can otherwise become an apparently certain statement in the summary.
A natural voice can sound particularly convincing. Apply the same scrutiny to its content as to written answers. Request a written overview when useful and check the underlying source. How confidently something is spoken does not change the evidence supporting it.
ChatGPT Work, agents and tools
ChatGPT Work remains a current feature at the review date. Official documentation describes it for tasks that produce completed work or a useful file, often across several steps and sources. A task might gather material, prepare a comparison and turn it into a presentation. The environment and permissions determine which actions are actually available.
This reflects the progression from a chatbot to an assistant with tools and then an agentic workflow. The system can organise steps and use intermediate results to continue. State the intended endpoint: advice, a new file or a change inside an existing application. These are materially different outcomes.
Connectors and plugins can connect the assistant to services or add capabilities. Access is not automatic. Installing a connection does not necessarily expose every file, and choosing a model does not grant account permissions. Specify the authoritative source, permitted data and when you want to review a draft.
An example shop task is to compare a product list with a supplied catalogue, flag discrepancies and prepare a correction file. Unknown values should remain unresolved, and publication should wait for review. This illustrates human-in-the-loop: the assistant performs preparation while a defined checkpoint controls the next action.
Current documentation also describes scheduled tasks and separate agent experiences. Access may depend on plan, region or rollout. A one-off request does not automatically create a recurring workflow. For repeated work, specify the source, update, timing and how missing information or failures should be reported.
Memory, context and personal information
AI Memory and the context window are different. Context is information available for the current processing step. Memory features can retain or retrieve useful preferences and earlier information, depending on product and settings. Neither means the assistant recalls every previous sentence perfectly.
For an ongoing project, use explicit working agreements and current source files. A style example, approved terminology and verified product details are useful anchors. When information changes, identify the current source clearly. A previous assistant response is not a substitute for an updated catalogue.
Review the data and memory settings for your account and workspace. Individual accounts, business workspaces and API applications do not automatically have identical conditions. Supply only what the task needs. In team work, establish who can access the sources and results. That access comes from the environment's configuration, not the language model.
Which ChatGPT plans currently exist?
| Plan | Who might use it? | What to check |
|---|---|---|
| Free | Trying the assistant, ordinary questions and occasional use. | Tools and more demanding features have more limited access. |
| Go | People using everyday features and uploads more frequently. | Expanded access compared with Free; the official page notes that ads may be included. |
| Plus | Regular individual work, research and file creation. | More access to advanced features, with usage limits still applying. |
| Pro | Intensive individual use and demanding assignments. | Several usage tiers exist within Pro; check the selected tier. |
| Business | Teams needing a shared workspace and administration. | The business page distinguishes Standard and Premium seats. |
| Enterprise | Organisations requiring more extensive administration and configuration. | Access, capacity and features follow the organisational arrangement. |
| Edu / education | Institutions with an appropriate educational use case. | Specific conditions and eligibility, rather than a universal individual consumer plan. |
Comparing plans on a consistent basis
Start with your working pattern. Do you need a few short answers, regular document creation, intensive research or collaboration? A higher plan is useful when its additional access or capacity matches that need. A team may require administration while an individual mainly needs longer sessions. Those are different reasons for selecting a plan.
Do not compare a monthly number in isolation. Check region, currency, tax, monthly or annual billing, seat count and possible additional credits. The official pricing page and final purchase interface are authoritative for those conditions. This entry deliberately avoids reproducing a figure that could be misunderstood without them. Separate dictionary articles for every plan are unnecessary.
Distinguish Chat from Work and Codex. Current model documentation places GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna in Work and Codex, not Chat. Specialist pricing documentation describes limited desktop access to Luna for Free and Go during rollout. The general plan table may show different model references. Where they differ, check the documentation for the particular feature and your account's model selector.
Work and Codex share usage according to current pricing documentation. API billing is separate. A per-token API rate therefore cannot be directly converted into the number of tasks included in a subscription. Likewise, unlimited wording should not be assumed to cover every feature. Read the relevant conditions and the usage display for the actual account.
Programming and a useful first request
ChatGPT can explain code and help prepare changes. For repository work, Codex is the relevant programming environment, bringing together code, a terminal, execution and review. A code response in a conversation is not yet a tested modification to your project. Ask what changed and which checks actually ran.
A useful first request identifies the goal, source, constraints and deliverable. For example: prepare three product-page drafts from these specifications, preserve every technical detail, flag missing information and return an editable file. Then check whether the result meets those requirements and refine the request where necessary.
Distinguish proposals from completed actions. Was a file actually created? Were sources opened? Was a calculation checked against the data? Were changes reviewed before publication? These questions make ChatGPT a practical aid without letting the conversational format imply more certainty than the available evidence supports.