ChatGPT 5: All Versions from GPT-5 to 5.6 Explained

👁 121 reads

If you’re searching for “ChatGPT 5,” you probably want to know one of two things: which version you’re currently using, or how the different numbered releases differ. The short answer to the first question is: on chatgpt.com you don’t choose a version - you get the current default model, and version numbers only matter when you’re working through the API or paying per token. The answer to the second question is in the table below, where we’ve compiled the entire GPT-5 family with dates, pricing, and context windows read from the model catalog on September 7, 2026.

All GPT-5 Versions in One Table

What This Means in Practice

Three things that only become clear when you line up the versions side by side.

Context window jumped once, in March 2026. The entire line from GPT-5 to 5.2 accepted 400 thousand tokens. From 5.4 onward it’s over a million, roughly two and a half thousand pages of text in a single query. If you work with long documents, this is the boundary that matters, not the incremental version number.

Pricing doesn’t grow linearly. GPT-5.5 cost $5 per million input tokens, while GPT-5.6 Sol released three months later costs just $2, with the same context window. Newer doesn’t mean more expensive. The exception is Pro variants, where output costs from $120 to $180 per million tokens - more than ten times the base version.

Luna is OpenAI’s cheapest viable model. $0.20 per million input tokens with a million-plus context window is a rate that changes the economics for automation processing large volumes of text. You can calculate the cost of specific tasks across models using our API cost calculator.

Sol, Terra and Luna: Why Three Variants of One Version

Instead of one model, OpenAI released three variants with different pricing profiles on July 9, 2026. Sol is the base variant and most commonly used, Terra has slightly more expensive output, Luna is the economy version. Each has a Pro counterpart. In practice the choice comes down to whether your task requires quality or throughput: Sol for conversations and writing, Luna for bulk processing. We covered the launch details and what happened around it in the GPT-5.6 hub.

What About ChatGPT 4?

Older models haven’t disappeared. Through the API you can still access GPT-4o ($2.50 per million input tokens), GPT-4.1 with a million-plus token window at $2, and classic GPT-4, which at $30 per million input tokens and an eight-thousand token window is now mainly a historical curiosity. In ChatGPT itself, OpenAI removes older models from the selection list when a successor appears.

Is there a reason to go back to GPT-4? Not really. Newer generations are cheaper with comparable or better results, and the difference in context window is dramatic: eight thousand tokens in GPT-4 versus over a million in the 5.4 line and newer. The only sensible reason is compatibility with legacy code that was written for a specific version and nobody wants to touch.

GPT-6 Astra: What Came After GPT-5

The GPT-6 family appeared in catalogs on September 4, 2026 at $10 per million input tokens and $50 for output. That’s five times the rate of GPT-5.6 Sol and OpenAI’s most expensive model in history, released in stages instead of one global launch. We’ve compiled everything about it, including our own tests, benchmarks and availability timeline, in the GPT-6 Astra guide, and you’ll find a direct comparison with the competition in Astra vs Claude Fable 5.1.

Which Version to Use for What

If you use ChatGPT in a browser or app, you don’t need to choose anything: OpenAI provides the current default model, and on paid plans you get a list with more powerful variants. If you’re working through the API, follow three questions. Does the task require long context? Then 5.4 or newer. Does cost matter with high query volume? Then Luna. Is it the hardest reasoning and can you pay for it? Then Pro variants or GPT-6 Astra. Current rates for all models, including competitors, are collected in the AI model ranking and model atlas.

See also: ChatGPT Guide: Plans, Pricing, Models and 20 Use Cases

If you use more than one model, my service Jantaris puts GPT-6 Sol, Claude Opus 5.5, Gemini 3.6 Flash, DeepSeek V4, Veo 3.1 and Nano Banana Pro on one prepaid USD balance with one API key - no subscription, the balance never expires and failed generations cost $0. Everything in one place: the Jantaris guide.

// read moreChatGPT Plus vs. OpenAI API: How Much Can $20 Buy?AI ToolsChatGPT Guide: Plans, Pricing, GPT-6 Models, and 20 UsesGuidesChatGPT 4: What Happened and What Replaced ItGuides

Kinetyra