Three assistants, one mechanism, different strengths.

How does ChatGPT work?

ChatGPT is a chat interface in front of OpenAI's language models. Your message, the conversation so far and a system prompt are turned into tokens and sent to the model, which generates a reply one token at a time, streamed back to the page. Around the model sit features: web search, file reading, code execution, image generation and memory of your preferences, each implemented as a tool the model can call or context the app adds. The models were trained on a vast text corpus to predict the next token and then tuned with human feedback to be helpful and safe. Claude and Gemini follow the same design with their own models and tools.

How ChatGPT works technically

The model is a decoder-only transformer with billions of parameters. Input tokens are embedded as vectors; dozens of layers of attention and feed-forward networks process them so that each position carries information from the whole context; the final layer outputs a probability distribution over the vocabulary; a token is sampled and appended; the loop repeats. Training has stages: pretraining on text, supervised fine-tuning on instruction examples, and preference optimisation (reinforcement learning from human feedback or related methods). Newer models add a reasoning stage that generates intermediate thinking before the answer. Serving uses batching, caching and specialised hardware to make generation fast for millions of users.

How do ChatGPT tokens work?

Tokens are the units the model reads and writes: pieces of words, about three-quarters of an English word each, more per word for other scripts. Every message you send and every reply counts in tokens, and the model's context window caps how many it can hold at once, so a long conversation eventually forgets its start. In the consumer app you do not pay per token, but usage limits on the free and paid plans are enforced in terms of messages or compute that come down to tokens. Through the API you pay per token for input and output, which is why concise prompts and caching matter for builders.

How do ChatGPT projects work?

Projects are folders inside ChatGPT that group conversations with shared files and custom instructions. You upload documents and write instructions once; every chat in the project starts with those files available and those instructions applied, so you do not repeat context. Under the hood the files are indexed and retrieved into the model's context as needed, a lightweight form of retrieval-augmented generation, and the instructions act as a project-level system prompt. Claude offers a similar feature with the same name, and Gemini's equivalents live in its Gems and in NotebookLM for document-centred work.

Which ChatGPT model is best for coding?

The model picker changes every few months, so check the current list: each provider now offers a general model and a reasoning model, and for coding the reasoning model usually wins on multi-step problems while the general model is faster for small edits. The same holds in Claude and Gemini. More important than the model is how you use it: give it the relevant files, state the constraints and the tests, ask for a plan before code on big changes, and read the diff. For serious work, coding agents that run in the terminal or editor and can execute tests outperform pasting into a chat window with any model.

ChatGPT vs Claude

Both are excellent general assistants and the gap on most everyday tasks is small; what differs is character and ecosystem. In my daily use Claude tends to write and reason in long, careful prose, handles long documents well and is strong at coding, especially through its agentic coding tool; ChatGPT has the broadest feature set, strong image generation, wide third-party integrations and the largest user base. Both have reasoning models, projects, file handling, web search and memory. Pricing tiers are similar. The right answer is to run your own thirty prompts through both free tiers and keep the one whose outputs you edit less.

What is Claude AI vs ChatGPT?

Claude is Anthropic's assistant, ChatGPT is OpenAI's. They are competitors built on the same kind of technology, and each company trains its own models with its own data, methods and priorities. Anthropic positions Claude around careful, honest behaviour and strong performance on writing, analysis and code; OpenAI positions ChatGPT as a broad consumer product with many modalities and integrations. Both offer APIs for developers, with Anthropic also publishing the Model Context Protocol that both ecosystems now support. For a student, having an account on each and comparing on real work is the quickest way to form your own view.

What is Claude AI good for?

• Long documents: reading, summarising and answering across large contexts.
• Writing that needs structure and restraint: reports, documentation, careful explanations.
• Coding, particularly multi-file work through its agentic tools.
• Analysis of data and arguments where you want it to show its reasoning.
• Building: its API and agent toolkit, and MCP support for connecting tools.
• Where it is weaker relative to rivals: image generation (it does not generate images) and the breadth of consumer integrations.

Can I use ChatGPT for free?

Yes. The free tier gives access to a capable default model with limits on the number of messages, file uploads and advanced features per period, and limited access to the top models and reasoning modes. Paid plans raise limits and unlock the strongest models and tools. For most students the free tier is enough for learning and everyday tasks; the paid tier becomes worth it when you hit limits daily or need the best model for coding and research. Claude and Gemini have the same shape: free with limits, paid for more.

How to get Claude AI for free

Sign up at Claude's website or install the mobile app; the free plan requires an account and gives daily message limits with the default model, file uploads and projects. There is no way to remove limits without a paid plan, and third-party sites promising unlimited free access are either reselling API usage against their terms or harvesting accounts; avoid them. For building, the API is pay-as-you-go with no permanent free tier, so for free experimentation use the chat app, or run open models locally with Ollama, or use providers that offer free API tiers for development, as I describe on the deployment page.