Skip to content
itsorted.online
AI guide

GPT-6 for Business: What Changed and What to Do (Oct 2026)

OpenAI's GPT-6 arrived in stages: GPT-6 Astra in early September 2026, GPT-6 Sol and Luna on 22 September, GPT-6.1 Sol on 29 September, and on 7 October GPT-6 became the default chat model for everyone. For a small business, the useful questions are which model your plan gets, what the API now costs, and whether anything you built on older models needs checking. This guide answers those from OpenAI's own pages as of October 2026. We have not run our own tests; benchmark figures are from published leaderboards with dates.

By IT Sorted teamReviewed 7 min readHow we write guides
in

Key takeaways

  • GPT-6 has three models: Astra (flagship), 6.1 Sol (mid-range) and Luna (small and cheap).
  • Since 7 Oct 2026, ChatGPT chat runs on GPT-6 Sol for Plus, Pro, Business and Enterprise, and Luna for Free and Go.
  • API list prices per million tokens: Astra US$10 in / US$50 out, 6.1 Sol US$2 / US$10, Luna US$0.10 / US$0.50.
  • ChatGPT Business stays at US$20 a seat on annual billing (US$25 monthly), with access to all GPT-6 models.
  • Re-test chatbot prompts and cost per reply before switching models; do not assume the new default is better for you.
In this guide · 7 sections
  1. 01GPT-6 timeline: what OpenAI released and when
  2. 02Which ChatGPT plans get which GPT-6 model
  3. 03GPT-6 API prices per million tokens
  4. 04How GPT-6 compares on published benchmarks
  5. 05What a small business should re-test now
  6. 06Do it yourself or get help?
  7. 07Frequently asked questions

Short answer

GPT-6 comes in three sizes: Astra (most capable, US$10/US$50 per million input/output tokens), 6.1 Sol (mid-range, US$2/US$10) and Luna (cheapest, US$0.10/US$0.50). In ChatGPT, paid plans including Business chat with GPT-6 Sol; Free and Go get Luna.

GPT-6 timeline: what OpenAI released and when

OpenAI did not launch GPT-6 in one go. The dates below are from OpenAI's announcement posts and its news feed.

DateReleaseWhat it means
3 Sep 2026GPT-6 AstraFlagship model, first for paid ChatGPT plans and the API
22 Sep 2026GPT-6 Sol and GPT-6 LunaMid-range and small models added to the family
29 Sep 2026GPT-6.1 SolUpgraded Sol at one-fifth of Astra's token price, in ChatGPT Work, Codex and the API
7 Oct 2026GPT-6 and Intelligent UI for everyoneGPT-6 becomes the chat model for all ChatGPT tiers

OpenAI's 7 October post says GPT-6 in ChatGPT is powered by GPT-6 Sol for Plus, Pro, Business and Enterprise, and GPT-6 Luna for Free and Go. It also says the models powering ChatGPT's Work and Codex modes did not change in that release; there, OpenAI had already made GPT-6 Astra and GPT-6.1 Sol available to paid plans.

The visible change in chat is what OpenAI calls Intelligent UI: answers can include charts, buttons, forms and small interactive tools, not just text. OpenAI also says GPT-6 can start answering while it is still reasoning.

Which ChatGPT plans get which GPT-6 model

From OpenAI's ChatGPT pricing and business pricing pages, as of October 2026, in US dollars before tax:

PlanPriceGPT-6 modelsBusiness use
FreeUS$0GPT-6 LunaNot suitable for company data
GoUS$8/monthGPT-6 LunaPersonal plan
PlusUS$20/monthAstra, 6.1 Sol, Sol and LunaPersonal plan, no admin console
ProFrom US$100/monthExpanded access, Pro reasoning on AstraPersonal plan
BusinessUS$20/seat/month annual, US$25 monthly; Premium seat US$100/US$125All GPT-6 models, flexible accessCompany workspace, 2 to 200 people
EnterpriseCustomAll GPT-6 modelsSCIM, data residency, custom retention

OpenAI says Business includes SSO, central billing, connectors to Google Workspace, Microsoft 365 and Slack, and no training on business data by default. Our ChatGPT Business price guide covers seat types, GST and minimum seats in detail.

GPT-6 API prices per million tokens

If you have a chatbot, quote generator or automation calling OpenAI's API, the model name and price matter more than the chat app. List prices from OpenAI's model documentation, standard processing, as of October 2026:

Model (API name)Input / 1M tokensOutput / 1M tokensOpenAI's suggested use
GPT-6 Astra (gpt-6-astra)US$10US$50Hardest reasoning and coding work
GPT-6.1 Sol (gpt-6.1-sol)US$2US$10Complex work at lower cost; cached input US$0.10
GPT-6 Luna (gpt-6-luna)US$0.10US$0.50High-volume focused tasks, such as extracting invoice fields or classifying requests
All three list a 1.05 million token context window and 128K maximum output.

For a typical website FAQ bot, the answer is usually Luna or Sol, not Astra. OpenAI's own GPT-6 model guide suggests matching model and reasoning effort to the task and measuring cost per successful task. Our LLM API pricing comparison works through cost per conversation and compares Anthropic and Google.

How GPT-6 compares on published benchmarks

IT Sorted has not run its own scored tests. The figures below are from Artificial Analysis, fetched on 18 October 2026, which publishes an Intelligence Index and the cost of running its test set.

Model (effort)Intelligence IndexCost per index task
Claude Opus 5.5 (max)58$5.98
GPT-6 Astra (max)53$3.26
GPT-6.1 Sol (max)52$0.72
GPT-6.1 Sol (high)50$0.32
GPT-6 Luna (max)38$0.07

Two points for a business reader. GPT-6.1 Sol scores close to Astra at a fraction of the cost, which matches OpenAI's own claim. And the top model overall on this index is not an OpenAI model, so if you are choosing a vendor rather than a model, read our ChatGPT vs Claude vs Gemini comparison.

What a small business should re-test now

If you only use ChatGPT for drafting and research, there is nothing to do except check that staff use the company workspace. If you have anything built on the API, run a short test before changing the model name in production.

  1. List what calls OpenAI

    Website chatbot, WhatsApp bot, email classifier, quote tool, Zapier or Make steps. Note the model name each one uses today.

  2. Collect 20 real questions or inputs

    Use real customer questions and documents, with personal data removed, plus a few tricky ones.

  3. Run them on the current model and one GPT-6 model

    Usually GPT-6 Luna or 6.1 Sol. Keep your prompt the same at first.

  4. Score accuracy, tone and length

    Mark each answer right, wrong or risky. Check prices, opening hours and policies against your source.

  5. Compare cost per reply

    Each API response reports input and output tokens. Multiply by the prices above to get the cost per reply.

  6. Switch only if it is better or cheaper

    Change the model name, keep the old one noted for rollback, and watch logs for a week.

Watch for longer answers: newer models often write more, which raises output costs and can annoy customers in a chat window. Set a maximum length in your instructions.

Do it yourself or get help?

Switching a model name is a one-line change; knowing whether answers got better or worse is the real work. If nobody on your side can run the test above, ask whoever built the integration.

IT Sorted is a Singapore-based team of developers with over 10 years of experience. AI Website Assistant (S$790) builds a chatbot on your own FAQ and price list, using the model that fits your volume and budget, and Workflow Automation (S$390 per workflow) connects AI steps to your email, forms and spreadsheets. You pay OpenAI directly for API use.

Frequently asked questions

They are three sizes of OpenAI's GPT-6 family. Astra is the most capable and most expensive, for hard reasoning and coding. GPT-6.1 Sol is the mid-range model that OpenAI says nearly matches Astra on many work tasks at one-fifth of the price. Luna is the small, cheap model for high-volume tasks such as classifying requests.

Yes. OpenAI's business pricing page lists GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna as available on Business with flexible access. Since 7 October 2026, the default chat model for Business is GPT-6 Sol. Business costs US$20 per seat per month on annual billing or US$25 monthly, as of October 2026.

As of October 2026, OpenAI lists GPT-6 Astra at US$10 per million input tokens and US$50 per million output tokens, GPT-6.1 Sol at US$2 and US$10, and GPT-6 Luna at US$0.10 and US$0.50, on standard processing. Cached input is cheaper. Fast and Ultrafast modes cost more per token.

Only after testing. Run 20 real customer questions through your current model and a GPT-6 model with the same instructions, check accuracy and tone, and compare the cost per reply from the token counts. Newer is not always better for a narrow FAQ bot, and longer answers can cost more.

That depends on the plan, not the model. OpenAI says ChatGPT Business and Enterprise and its API do not train on business data by default. Free and personal plans are different. For PDPA purposes, keep company work in a business workspace or the API, and do not paste NRIC numbers or medical details into any AI tool.

Rather have it done for you?

Fixed price, set up in your own accounts.

Order online, get a written scope, and keep full ownership of everything we set up. Questions first? Ask us in writing.