ResourceAI

OpenAI’s GPT models in October 2026: which one to use in ChatGPT and the API

Astra, Sol or Luna? Which OpenAI GPT models to use in ChatGPT and the API as of October 2026, what they cost and which ones are going away.

Advertisement

Updated October 9, 2026.

This page replaces our October 2025 guide, which still covered GPT-3.5, GPT-4o and the o1, o3 and o4-mini models. OpenAI has since rebuilt its lineup. GPT-6 Astra came out on September 3, 2026, GPT-6 Sol and GPT-6 Luna on September 22, and GPT-6.1 Sol on September 29. GPT-6 came to everyday ChatGPT conversations on October 7. On October 12, Meta compared its new model with GPT-6 Astra, and our article on Meta’s Muse Wonder walks through the results. Google had already compared Gemini 4 Argon with GPT-6 Astra when it announced the model on September 30.

Everything here comes from OpenAI pages we checked on October 9, 2026, including the API model catalog, the ChatGPT pricing page, the help article on GPT-6 in ChatGPT and the model deprecations page. OpenAI lists its prices in US dollars. For readers in Europe, we add euro amounts at the European Central Bank reference rate for October 8, 2026 ($1.1186 to the euro).

Advertisement

Three tiers: Astra, Sol and Luna

The GPT-6 family comes in three tiers, from most capable to cheapest.

OpenAI describes GPT-6 Astra as its most capable model. It is built for long, demanding work: reasoning, coding, operating a computer, research and document creation.

GPT-6.1 Sol comes close to Astra’s results at a lower price, according to OpenAI. It succeeds GPT-6 Sol, which is still available.

GPT-6 Luna is the cheapest of the three and, according to OpenAI, the fastest. The company recommends it for narrow tasks you run at high volume, where you already know what a good result looks like: extraction, classification, rewriting text into another format, structured summaries. OpenAI’s Decisions API, generally available since October 22, 2026, runs gpt-6-luna on closed questions and doesn’t charge for output tokens.

Advertisement

Sol and Luna are not new names. The previous generation, GPT-5.6, released on July 9, 2026, also had three tiers: Sol, Terra and Luna. OpenAI’s model pages say Terra roughly matches the old “mini” tier and Luna the old “nano” tier. GPT-5.6 Terra’s successor, GPT-6.1 Terra, came out on October 26, 2026, and OpenAI prices it at half the rate of GPT-6.1 Sol.

In ChatGPT: what each plan gets you

In a regular chat, the model menu now shows “GPT-6” where it used to say “Latest.” A thinking slider sets how hard the model works: Instant for everyday questions, then Medium, High, Extra High, and Pro for the hardest tasks. On Plus, Pro, Business and Enterprise, the Instant through Extra High levels run on GPT-6 Sol. The Pro level uses GPT-6 Pro, which is built on GPT-6 Astra.

GPT-6 answers can also include visuals and interactive elements, such as a side-by-side comparison, a diagram or a small calculator. OpenAI calls this Intelligent UI. It isn’t available at the Pro level.

Plan Monthly price Chat model Thinking levels
Free $0 (€0) GPT-6 Luna (rolling out since October 8) Instant, plus Think for harder questions
Go $8 (€7.15) GPT-6 Luna (rolling out since October 8) Instant, plus Think for harder questions
Plus $20 (€17.88) GPT-6 Sol Instant, Medium, High
Pro $100, $200 or $500 (€89.40, €178.79 or €446.99) GPT-6 Sol, GPT-6 Pro All, including Pro
Business, 2 users minimum $20 per user billed annually, $25 billed monthly (€17.88 or €22.35) GPT-6 Sol, GPT-6 Pro All, including Pro

The euro figures are our conversions. What you actually pay, tax and currency included, is the amount ChatGPT shows you at checkout.

Advertisement

The $500 Pro tier (€446.99) adds Ultrafast mode in Work and Codex. Astra and GPT-6.1 Sol respond faster there, but they use up your allowance eight times as fast.

Enterprise pricing goes through OpenAI’s sales team. On Business and Enterprise, workspace admins can choose which models members have access to.

The context window, meaning how much text ChatGPT can work with in one conversation, also depends on your plan. It is measured in tokens (small chunks of words). In Instant mode, the window is 27,000 tokens on Free (about 12 pages of input text), 54,000 on Go and Plus (about 40 pages) and 128,000 on Pro (about 250 pages). With reasoning, it grows to 256,000 tokens on Go and Plus and to 400,000 on Pro.

ChatGPT Work, which takes a task all the way to a result you can review, and Codex, OpenAI’s coding tool, have their own model choices. Plus subscribers get GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in both. The GPT-6 launch in chat doesn’t change anything there.

Advertisement

Usage is changing as well: starting in December, ChatGPT and Codex will share one usage limit counted in tokens on Plus, Pro and Business.

GPT-5.5 leaves ChatGPT, Work and Codex on October 14, 2026, on every plan. It stays available in the API. Custom GPTs follow on December 11, 2026: see the ChatGPT GPTs retirement timeline and the move to plugins.

In the API: which model for which job

The API is how your own software talks to OpenAI’s models. The prices below are standard rates per million tokens, for requests of up to 272,000 input tokens.

Use case Model Input Output
Long, hard tasks: agents, complex code, research, documents GPT-6 Astra $10 (€8.94) $50 (€44.70)
Recurring complex work where cost matters GPT-6.1 Sol $2 (€1.79) $10 (€8.94)
High volume: extraction, classification, structured summaries GPT-6 Luna $0.10 (€0.09) $0.50 (€0.45)
Testing ChatGPT’s Instant model (not for production) chat-latest $5 (€4.47) $30 (€26.82)

All three GPT-6 models read text and images and reply in text. Their context window is 1,050,000 tokens, with up to 922,000 for input and 128,000 for output. Once a request goes past 272,000 input tokens, the whole request costs twice as much for input and one and a half times as much for output.

Batch and Flex, the two deferred processing options, cost half the standard rate, and Fast mode costs twice as much. Cached input tokens cost far less: $1 (€0.89) per million for Astra and $0.10 (€0.09) for GPT-6.1 Sol.

If you followed our old guide, two things have changed. Our 2025 guide told you to tune temperature (0.2 for a spec sheet, 0.8 for brainstorming), but GPT-6 Astra no longer accepts custom temperature or top_p values. Astra also needs the Responses API to call tools (your own functions, web search, code execution).

Data residency, meaning where your requests are processed, matters for European companies. GPT-6.1 Sol can process your requests in Europe (the EEA and Switzerland), including in Fast and Ultrafast modes. GPT-6 Sol and GPT-6 Luna offer European processing with Standard, Fast, Flex and Batch.

Advertisement

GPT-6 Astra requests can also be processed in Europe, except in Fast and Ultrafast modes. Regional processing adds 10% to the price.

OpenAI also offers specialized models. GPT-Live 1, generally available since September 10, 2026, carries on a continuous spoken conversation. It costs $0.05 a minute (about 4.5 euro cents), billed by the second, plus the cost of whatever model and tools it calls. GPT Image 2.5 Sunburst and GPT Image 2.5 Flare, released on September 8, create and edit images.

If you want to run a model on your own servers, OpenAI publishes gpt-oss-120b and gpt-oss-20b under the Apache 2.0 license. The Agents API, for building agents, has been in public beta since September 10.

Retired and soon-to-be-retired API models

GPT-4.5 Preview, which our 2025 version still listed, had already shut down on July 14, 2025. On October 23, 2026, the API will turn off GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, GPT-4.1 nano, o1, o1-pro, o3-mini and o4-mini. OpenAI suggests GPT-5.6 Sol, Terra or Luna as replacements, depending on the model.

The August 2025 versions of GPT-5, GPT-5 mini and GPT-5 nano follow on December 11, 2026, along with GPT-5 Pro, o3 and o3-pro. GPT-4.1, GPT-4.1 mini and the current version of GPT-4o are still in the catalog, with no shutdown date announced. If any of your company’s software calls a model on these lists, plan your migration before those dates.

Limits and things to watch

According to OpenAI’s October 7 announcement, the Free and Go plans move to GPT-6 Luna starting October 8. OpenAI’s help article, when we read it on October 9, still listed GPT-5.6 Luna for those plans. The model menu in your account shows what you actually have.

For most levels, OpenAI doesn’t publish a fixed number of messages. Limits depend on your plan, the model and the size of the task. When you hit one, ChatGPT shows the reset time if it has it. Even the Pro plan doesn’t include unlimited use of GPT-6 Pro.

Advertisement

The number of subscriptions has limits too: ChatGPT can block an extra paid subscription on the same payment card or device, according to a help page updated October 30.

GPT-5.6 Sol currently costs $4 (€3.58) per million input tokens and $20 (€17.88) per million output tokens in the API. That is a promotional rate, which OpenAI says will last at least through November 21, 2026.

The performance figures published at each launch are OpenAI’s own measurements. The company itself suggests comparing GPT-6.1 Sol and Astra on your own tasks before you choose.

“GPT-6” is not an API model ID. Behind that name, ChatGPT runs GPT-6 Sol on Plus, Pro, Business and Enterprise, and GPT-6 Luna on Free and Go, in versions OpenAI says are tuned for everyday conversation. Keep that in mind if you compare a ChatGPT answer with one you got from gpt-6-sol or gpt-6-luna through the API.

Where to start

For office work, stick with the plan you have and set the slider to match the task: Instant for an email, Medium or High to analyze a document if your plan includes them (Plus and up). For a software project, try GPT-6.1 Sol before Astra, and Luna for high-volume jobs.

If what you mostly need is help with everyday paperwork, like follow-ups, quotes and invoices, the MACCUS AI assistant handles it for you. You message it on Telegram, by text or voice note.

Advertisement

Leave a Comment