GPT-6 explained: Astra, Sol, Terra and Luna

GPT-6 explained for business users: what OpenAI's Astra, Sol, Terra and Luna models are for, when each arrived, API prices and which one to use.

Updated3 November 2026
Reading time4 min read

GPT-6 is OpenAI’s current family of models, released from September 2026. It has four tiers: Astra at the top, then Sol, Terra and Luna. This guide explains what each GPT-6 model is for, when it arrived, what it costs and which to use.

What each GPT-6 model is for

Each tier has a different job:

The Sol, Terra and Luna names first appeared with GPT-5.6 in July. The model names gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna date from 9 July 2026. Astra is new with GPT-6.

For help choosing between models and tuning reasoning effort, read OpenAI’s model guide for the GPT-6 family.

When each model was released

Model Announced Model name
GPT-6 Astra 3 September 2026 gpt-6-astra
GPT-6 Sol 22 September 2026 gpt-6-sol
GPT-6 Luna 22 September 2026 gpt-6-luna
GPT-6.1 Sol 29 September 2026 gpt-6.1-sol
GPT-6.1 Terra 26 October 2026 gpt-6.1-terra

Dates are from OpenAI’s announcements. Model names and versions are as listed in Microsoft’s Azure OpenAI documentation.

How much GPT-6 costs in the API

GPT-6.1 Sol costs one-fifth of Astra’s standard API input and output token prices.

The table below shows Microsoft Foundry’s Global Standard prices for the same models. Prices are in US dollars per million tokens, for short prompts.

Model Input Cached input Output
GPT-6 Astra US$10 US$1 US$50
GPT-6 Sol US$2 US$0.20 US$10
GPT-6.1 Sol US$2 US$0.10 US$10

Long prompts cost more. On Microsoft Foundry, long-context input is US$20 per million tokens for Astra and US$4 for Sol. Check OpenAI’s own pricing page before you budget for Terra or Luna, and for any change since this was written.

GPT-6 also brought better prompt caching, with explicit cache breakpoints. Caching cuts the cost of sending the same instructions or documents many times.

Where GPT-6 Astra stands on safety

GPT-6 Astra is OpenAI’s most capable broadly deployed model. It is also the first OpenAI model to reach the “Critical” level of cybersecurity capability under the company’s Preparedness Framework.

What changed in ChatGPT

Plans and usage limits change often. Check OpenAI’s pricing page for which models your ChatGPT plan includes.

Which GPT-6 model to use

  • Luna for sorting, tagging, extraction and short replies at high volume.
  • Terra for agents that operate websites and software.
  • Sol for coding, research and multi-step analysis.
  • Astra for the hardest problems, where accuracy matters more than cost.

Start low, move up only when needed

Overview
  1. GPT-6 LunaSorting, tagging, extraction and short replies at high volume
  2. GPT-6 TerraAgents that operate websites and software
  3. GPT-6 SolCoding, research and multi-step analysis
  4. GPT-6 AstraThe hardest problems, where accuracy matters more than cost

Start with the cheapest model that does the job well. Move up only for the steps that need it. Test each choice on 20 or so real examples from your own work before you commit.

For a comparison with Claude and Gemini, read Claude, GPT-6 or Gemini 4? Choosing an AI model. For the month’s wider news, read October 2026 AI updates.

References

References16 official sources

Get your business mentioned on AI assistants.