ChatGPT Features and Comparison with Competitors

First of all, OpenAI is about the only one using this unfamiliar string "GPT" in a product name. This GPT stands for Generative Pre-trained Transformer, which is basically another name for LLMs (Large Language Models). Since general users don't need to worry about this, it's best to understand it in terms of the advantages and challenges of being an industry pioneer.

CEO Sam Altman is a celebrity in the industry. He was once ousted? And then he came back (just like Steve Jobs at Apple, haha). Since it was initially established as an NPO (nonprofit organization), it was involved in various troubles. Even after establishing the for-profit entity, problems didn't cease, but it feels like things have finally settled down recently. However, partly due to the drop in motivation among some staff because of the shift to a for-profit focus midway, they are currently trying to overcome this through a labor-intensive approach of increasing staff size using their vast funding.

Now, as for OpenAI's ChatGPT, because it has a longer history than other tools, its natural conversational responses are probably the best in the industry. Currently, if we rank the conversational abilities in the AI industry, it would look something like this:

Natural Conversational Ability: OpenAI ChatGPT > Claude Code = Google Gemini

<Conversational Characteristics of Each Product>

Product	Conversational Characteristics
OpenAI ChatGPT	Extremely casual. Especially if the prompt (user instructions) is in spoken style, it becomes very informal. That's why in previous versions, many people reportedly started treating ChatGPT like a friend.
Claude Code (claude.ai)	Flat, without much emotion, giving textbook-like responses. It feels very good for programmers.
Google Gemini	Incredibly polite. It treats the user as a superior, always seeking approval as if it's a salesperson pitching to a client. Personally, I feel it's a bit too much.

Though, when it comes to performing complex and structured tasks like programming, the evaluations differ.

Programming & Tasks: Claude Code > ChatGPT (codex) >> Google Gemini

<Coding Capability Characteristics of Each Product>

Product	Characteristics
ChatGPT (codex)	It was not very good up to 5.3, but has improved significantly with the current 5.4. However, it often overthinks and goes in the wrong direction, so you get better results by specifying the direction to delve into. It's about a mid-level programmer.
Claude Code	Currently the best in the world for programming. It reportedly handled many operations, unit actions, and investigations at the US Department of Defense, contributing to massive savings in time, personnel, and costs. It has the highest stability in actual coding. Since it is specialized in programming, it acts on its own judgment to some extent without explanation, but it can sometimes run off in the wrong direction, so caution is needed. It's about an senior programmer.
Google Gemini	Google is investing massive staff and budget to catch up. It improved slightly in Gemini 3, but there are many things Gemini cannot solve that the two AI models above can, and it feels like it lacks training data. It's about a junior programmer.

<About ChatGPT Models>

There are many model options in ChatGPT, but honestly, you can only entrust simple tasks to anything other than the latest model. So for some people, the unnecessarily large number of choices might just be in the way.

Model Name
ChatGPT 5.4, ChatGPT 5.4 mini	The latest 5.4 series. Capable of more complex reasoning, but takes longer. The mini is a lightweight version focused on response speed.
ChatGPT 5.3 codex	The top-tier model specialized in programming.
ChatGPT 5.2, ChatGPT 5.2 codex	Already legacy. You can only ask for really simple things. But the conversation is great, haha.
ChatGPT 5.1 codex-max, ChatGPT 5.1 codex-mini	Probably won't be used anymore. Completely low-tier models.

<About ChatGPT Reasoning Options>

In ChatGPT, besides the model, you can specify the depth of reasoning. This seems to create a big difference in reasoning even for the same model.

Reasoning Level	Characteristics
Low	Fastest, answers almost intuitively without deep validation.
Medium	Balanced (default), includes light reasoning.
High	Thinks in multiple steps, reduces mistakes, slightly slower. Much stronger at reasoning than the balanced level, likely because it breaks down the problem.
XHigh	Thinks deeply over a long time, using a hypothesis-verification-correction flow. Most accurate but slowest. It often finds its own mistakes during validation.

<Common Settings and Exclusions in ChatGPT>

You can set these by going to your name in the bottom-left of the ChatGPT screen > Settings > Personalization > Custom Instructions. Since these settings apply to the entire thread with ChatGPT, they are suitable for common settings or parameters you want to specify every time. In the case of CLI (terminal use), there is no explicit way to specify an external file like Claude Code does (CLAUDE.md), but it tries to create files named AGENTS.md or README.md.

<Description of Plans>

Service Provided	Price	Models Used	Overview
ChatGPT Free	Free	GPT-5.2 DALLE-	Latest models like GPT-5.2 are limited to about 10 messages per 5 hours. A single Q&A is around a few thousand tokens (several thousand Japanese characters). Moves to the GPT-4o mini model after reaching the limit. Image generation also has separate limits.
ChatGPT Go	$8/month	GPT-5.2 Instant	10 times the messages of the free plan, file uploads, and image generation available.
ChatGPT Plus	$20/month	GPT-5.2 Thinking Codex	Access to advanced models is expanded further. Legacy models are selectable, and the coding agent Codex is also available. Designed for tasks requiring deeper reasoning like content creation, document writing, learning, research, and data analysis.
ChatGPT Pro	$200/month	GPT-5.2 Pro Codex	Top-tier plan for power users. Virtually unlimited messages, uploads, and image generation. Guarantees the fastest and highest performance of top models.

<Codex>

An AI specialized in coding assistance by OpenAI. Many people use it alongside Claude Code. While there is a standalone Codex service, a subscription format as a feature of ChatGPT Plus is also popular. Even in that case, it is $20/month, so many developers use it in combination with Claude Code, switching between them depending on usage limits.

<Free Plan>

As of February 2026, ChatGPT's free plan allows you to use the latest GPT-4o about 10–30 times every 5 hours. Once you hit the limit, it automatically switches to the lightweight version (mini) or applies a time limit. Image generation is limited to 3 images per day, and speeds can drop significantly during peak times.