GPT models (OpenAI)

What it is

OpenAI’s GPT family is the model line behind ChatGPT and the OpenAI API. The current flagship tier has three steps (Astra, Sol, Luna). Separate specialised models cover images, voice, transcription and security research.

How it works

On the API you call a model through the Responses API and set a reasoning effort. Higher effort means more thinking, more tokens and more cost. Built-in tools let the model search the web, search your uploaded files, call your functions, or operate a computer interface. In ChatGPT, the same capabilities appear as features rather than settings.

Specs and pricing (checked 2026-10-17). All three models list a 1.05M-token context and up to 128K output tokens. API prices per million input and output tokens are $10 and $50 for Astra, $2 and $10 for Sol, and $0.10 and $0.50 for Luna. ChatGPT plans run from Free through Go ($8 a month) and Plus ($20 a month) to Pro, from $100 a month.

Benefits

• Three clear cost steps with the same context window, so you can drop a tier without redesigning prompts.
• Built-in tools reduce the glue code needed for search and file Q&A.
• Connects to other tools. OpenAI’s developer docs describe adding a custom MCP server in ChatGPT through its Plugins page, subject to account and workspace policies (see the MCP introduction).
• Packaged instructions. Custom GPTs turn a repeatable prompt into a shared tool (see SOPs as custom assistants).

Limits and risks

• Effort costs money. Christopher Penn’s example is a viewer who spent heavily on a top reasoning tier for simple documentation work. Match the effort setting to the task.
• Fast-moving lineup. Names, prices and cutoffs here are a dated snapshot.
• Knowledge cutoff. Without web search turned on, answers reflect training data up to the cutoff (see knowledge cutoff).