ChatGPT Models Explained: GPT-6 Astra, GPT-5.6 Sol and Luna, and Which Model Is Best

The latest ChatGPT model is GPT-6 Astra, which OpenAI introduced on September 3, 2026. In regular ChatGPT chats it appears as GPT-6 Pro on the Pro ($100, $200 and $500), Business and Enterprise plans, while Plus gets Astra inside ChatGPT Work and Codex. Everyday chats run GPT-5.6 Sol on paid plans and GPT-5.6 Luna on Free and Go (OpenAI Help Center, checked September 25, 2026). Older models, including GPT-4o, GPT-4.1, o4-mini, GPT-5.1, GPT-5.2 and GPT-4.5, have been retired from ChatGPT during 2026.

The ChatGPT model picker changed three times in the last six weeks: GPT-5.6 Sol and Luna replaced GPT-5.5 Instant on August 6, GPT-6 Astra arrived on September 3, and on September 14 OpenAI stopped Plus and Pro chats from switching automatically from Instant to a thinking level. This guide explains what each current model is, which plan can use it, how the Instant, Medium, High, Extra High and Pro options map to real models, and which model is the best pick for your task and your plan. Every figure below comes from OpenAI's own help articles, pricing page, release notes and developer documentation, linked in the sources.

By Adi Leviim, Co-Founder at Infi Developments. Last reviewed September 25, 2026.

What Is the Latest Version of ChatGPT?

As of September 15, 2026, the latest version of ChatGPT is GPT-6 Astra, introduced by OpenAI on September 3, 2026 and rolled out to paid plans over the following days. OpenAI calls it its most capable model and describes it as state of the art on computer use, browsing, software engineering, cybersecurity, science and professional work. The previous flagship, GPT-5.6 Sol, launched on July 9, 2026 and still powers most paid chats.

Which model actually answers you depends on your plan, because OpenAI now exposes models through a set of thinking levels rather than a list of model names:

Source: GPT-5.6 and GPT-6 Pro in ChatGPT (OpenAI Help Center, updated September 15, 2026). Enterprise workspaces may still see the older model picker until the new controls reach them, and in Enterprise and Edu workspaces Astra is off by default until an owner enables it.

Which ChatGPT Model Is Best? How to Choose by Task and by Plan

There is no single best ChatGPT model: GPT-6 Astra is the most capable, but for most tasks the best choice is GPT-5.6 Sol at the lowest thinking level that gets the answer right, because higher levels are slower and draw on limited allowances. Use Instant for quick answers, High for coding, math and data analysis, and save GPT-6 Pro for the hardest multi-step problems. On Free and Go there is only one model, GPT-5.6 Luna, so the choice is between a normal reply and the Think button.

Best ChatGPT model by task

The recommendations below follow OpenAI's own descriptions of each option: Instant for "everyday questions", Medium for standard thinking, High for extended thinking, Extra High as the highest Sol level, and Pro for "difficult tasks and longer-running workflows."

Best ChatGPT model by plan

• Free and Go: GPT-5.6 Luna is the only chat model and text chats are unlimited, subject to abuse guardrails. Press Think for harder questions; it still uses Luna, not Sol.
• Plus ($20/month): GPT-5.6 Sol. Stay on Instant for everyday work, switch to Medium or High yourself for hard problems, because Plus chats no longer switch up automatically. Use GPT-6 Astra in ChatGPT Work or Codex when you want a finished deliverable or a long coding task.
• Pro ($100, $200 or $500/month): GPT-5.6 Sol up to Extra High for most hard work, and GPT-6 Pro for the hardest problems. OpenAI no longer publishes per-tier message counts: Pro plans with higher allowances get higher Pro-model limits, and on Pro $100 GPT-6 Pro and GPT-5.6 Sol Pro share one weekly allowance.
• Business: GPT-5.6 Sol for daily work. GPT-6 Pro and GPT-5.6 Sol Pro share 15 messages per month on Standard seats and 50 per week on Premium seats, so reserve them.
• Enterprise and Edu: whatever your admin enables. GPT-6 Astra is off by default at launch and must be turned on per workspace or role.

If you are choosing a plan rather than a model, the context window matters as much as the model name: see ChatGPT context window by plan and is ChatGPT Plus worth it.

The 2026 ChatGPT Lineup: GPT-6 Astra and the GPT-5.6 Family

ChatGPT's current lineup in September 2026 is GPT-6 Astra at the top, then GPT-6 Sol and GPT-6 Luna in ChatGPT Work and Codex, then the three GPT-5.6 tiers: Sol (flagship), Terra (balanced) and Luna (fastest and cheapest). In ordinary ChatGPT conversations you meet only three of them: Luna on Free and Go, Sol on paid plans, and GPT-6 Astra as GPT-6 Pro. OpenAI states that Terra and Luna are not selectable in standard ChatGPT conversations on paid plans; they are available in ChatGPT Work, Codex and the API.

• GPT-6 Astra: newest and most capable; GPT-6 Pro in Chat on Pro, Business and Enterprise; in Work and Codex on Plus and above
• GPT-6 Sol and GPT-6 Luna: added on September 22, 2026 for ChatGPT Work and Codex only. OpenAI's help centre states that they "are not available in regular ChatGPT conversations" and that availability depends on your plan, workspace settings and rollout access
• GPT-5.6 Sol: powers Instant, Medium, High and Extra High on paid plans
• GPT-5.6 Sol Pro: the Pro option of the GPT-5.6 family on Pro, Business and Enterprise
• GPT-5.6 Terra: balanced model in ChatGPT Work and Codex (Codex on Free and Go uses Terra)
• GPT-5.6 Luna: default chat model on Free and Go, and in Work and Codex on paid plans

OpenAI's pricing page also lists legacy models as included on Plus and Pro, and its help article notes that earlier models "may remain available where supported," but OpenAI does not publish which older models remain in the legacy menu for individual plans.

Can I still use GPT-4, GPT-4o or GPT-4.5 in ChatGPT?

No. These models have been retired from ChatGPT, though several remain in the OpenAI API. The ChatGPT retirement dates OpenAI has published:

If a 2025 article recommends GPT-4o, o3 or GPT-4.5, it no longer describes ChatGPT's model picker. Our GPT-4.5 retirement guide covers that model's history.

GPT-6 Astra (GPT-6 Pro): OpenAI's Newest Flagship

GPT-6 Astra is OpenAI's most capable model, introduced on September 3, 2026, and in ChatGPT it is reached through the GPT-6 Pro option on Pro, Business and Enterprise plans or inside ChatGPT Work and Codex. OpenAI first rolled it out to a limited set of organizations, then to Plus, Pro, Business and Enterprise users over the following days, and to the API, Microsoft Azure and AWS Bedrock.

• In Chat: GPT-6 Pro on Pro $100, Pro $200, Business and Enterprise (OpenAI's GPT-6 Pro article lists these plans); not on Free, Go or Plus
• In Work and Codex: Plus and Business Standard seats get limited Astra usage with optional credits; Pro and Business Premium can use their full Work and Codex allowance
• Chat allowances: OpenAI no longer publishes per-tier message counts for Pro; plans with higher allowances get higher Pro-model limits, and Pro $100 shares one weekly allowance across GPT-6 Pro and GPT-5.6 Sol Pro; Business Standard, 15 per month shared; Business Premium, 50 per week shared
• When the allowance runs out: on Pro $200, ChatGPT switches to GPT-5.6 Thinking at Medium; you can still pick GPT-5.6 Sol Pro while its own daily allowance lasts
• API model: gpt-6-astra, 1,050,000-token context window, 922,000 maximum input tokens, 128,000 maximum output tokens, knowledge cutoff April 30, 2026
• API price: $10 per million input tokens, $50 per million output tokens, $1 cached input; prompts above 272,000 input tokens bill at 2x input and 1.5x output

OpenAI's published results give a sense of the jump over GPT-5.6 Sol: 57.9% on Terminal-Bench 4.0 against 37.3% for Sol, 96.0% on GPQA Diamond, and 72.6% on OSWorld 2.0 in about 40 minutes per task against Sol's 65.7% in about 75 minutes. OpenAI also rates Astra at the Critical threshold for cybersecurity under its Preparedness Framework, so it refuses more advanced tasks such as building proof-of-concept exploits, and some conversations may pause for your review. Details are in the GPT-6 Astra announcement; benchmark scores are OpenAI's own, run in its research environment.

GPT-5.6 Sol: The Model Behind Instant, Medium, High and Extra High

GPT-5.6 Sol is the flagship of the GPT-5.6 family and the model that answers almost every chat on Plus, Pro, Business and Enterprise, from Instant up to Extra High. It launched on July 9, 2026, and on August 6, 2026 OpenAI shipped an updated Sol for ChatGPT with "more reliable facts, more focused answers," and a slider for how much thought ChatGPT puts into a reply.

• Instant: fast responses for everyday questions
• Medium: standard thinking (Plus, Pro, Business, Enterprise)
• High: extended thinking (Plus, Pro, Business, Enterprise)
• Extra High: the highest Sol thinking level (Pro, Business, Enterprise; not Plus)
• Pro: GPT-5.6 Sol Pro or GPT-6 Pro (Pro, Business, Enterprise; not Plus)
• API model: gpt-5.6-sol (alias gpt-5.6), 1,050,000-token context window, 128,000 output tokens, knowledge cutoff February 16, 2026

Since September 14, 2026, Instant on Plus and Pro no longer switches automatically to a higher thinking level for complex requests, and asking ChatGPT to "think harder" does not change the level. ChatGPT can still switch for safety reasons. If a hard question gets a shallow answer, pick Medium or High yourself. How much you can paste into one message also depends on the model and plan; see the ChatGPT character and word limit guide.

GPT-5.6 Luna: The Default for Free and Go

GPT-5.6 Luna has been the default ChatGPT model for Free and Go users since the week of August 6, 2026, when it replaced GPT-5.5 Instant, and both plans now have unlimited everyday text chats. Luna is the fastest and lowest-cost model in the GPT-5.6 family.

• Availability: default and only chat model on Free and Go; also in ChatGPT Work and Codex on paid plans
• Text chats: unlimited on Free and Go, subject to abuse-prevention safeguards
• Still limited: file uploads, image generation, voice, data analysis and other tools
• Think button: a deeper reasoning pass for harder questions, powered by Luna, not Sol
• No Sol access: Free, Go and logged-out users cannot use GPT-5.6 Sol in Chat
• API price: $0.20 per million input tokens and $1.20 per million output tokens

For a free user, the practical upgrade over the old default is the removal of the message cap on text chats. What Free still lacks is the larger context window: 27K tokens for instant replies against 54K on Go and Plus.

GPT-5.6 Sol, Terra, and Luna: The Frontier Family Compared

All three GPT-5.6 tiers share the same limits in the API, a 1,050,000-token context window, 922,000 maximum input tokens and 128,000 output tokens, and differ in capability, speed and price. OpenAI's naming puts the generation in the number and the tier in the name: Sol roughly corresponds to the unsuffixed model of earlier GPT-5 families, Terra to the mini tier, and Luna to the nano tier.

API prices dropped after launch: GPT-5.6 launched at $5 / $30 for Sol, $2.50 / $15 for Terra and $1 / $6 for Luna, then OpenAI cut Luna by 80% and Terra by 20% on July 30, 2026, and Sol by more than 20% on August 21, 2026. Figures are from OpenAI's model documentation. Note that these API windows are not what you get in the ChatGPT app, where the window is set by your plan (next section).

GPT-5.5 and GPT-5.4: The Previous Flagships

GPT-5.5 and GPT-5.4 are the two model generations before GPT-5.6; both remain available in the OpenAI API, but in ChatGPT they have been replaced by GPT-5.6 Sol and Luna as the models behind everyday chats. Earlier models may remain in the legacy menu where supported, and OpenAI does not list which ones for individual plans.

GPT-5.5 and GPT-5.5 Pro

GPT-5.5 Instant became ChatGPT's default for all users on May 5, 2026, replacing GPT-5.3 Instant, and on May 28, 2026 OpenAI removed canvas from GPT-5.5 Instant and GPT-5.5 Thinking in favor of writing and code blocks in the chat. GPT-5.5 Instant was then replaced by GPT-5.6 Sol for paid users and GPT-5.6 Luna for Free and Go in August 2026. In the API, GPT-5.5 (snapshot dated April 23, 2026) and GPT-5.5 Pro are still listed.

GPT-5.4 Thinking and GPT-5.4 Pro

GPT-5.4 Thinking launched in ChatGPT on March 5, 2026 for Plus, Team and Pro users, replacing GPT-5.2 Thinking, with GPT-5.4 Pro on Pro and Enterprise. OpenAI reported that GPT-5.4's individual claims were 33% less likely to be false and its full responses 18% less likely to contain errors than GPT-5.2, and that it matched or beat industry professionals in 83.0% of GDPval comparisons (OpenAI, March 2026). GPT-5.4 mini served as the Thinking fallback and as the model behind the Free and Go Thinking feature from March 18, 2026, before Luna took over that role.

ChatGPT Version History: From GPT-3.5 to GPT-6 Astra

ChatGPT has run on more than a dozen model versions since its November 2022 launch on GPT-3.5, and every model before the GPT-5.5 generation has now been retired from the ChatGPT app. Here is where each version you might be searching for landed:

Looking for a specific older version? We keep guides for the ones people still search for: the GPT-3.5 model overview, GPT-4 vs ChatGPT Free, and the GPT-4.5 retirement guide.

Thinking Levels: Instant, Medium, High, Extra High and Pro

ChatGPT's model picker no longer lists model names for most users; it lists thinking levels, and on paid plans every level except Pro runs GPT-5.6 Sol. A higher level means more internal reasoning before the reply appears, which is why answers take longer; how ChatGPT works explains what happens in between. OpenAI renamed the old options on June 10, 2026, so if you remember Standard, Extended, Light or Heavy thinking, this is where they went:

On the web the picker sits in the message composer; on iOS and Android it is at the top of the conversation. The "Higher intelligence" setting that let Instant add reasoning automatically was removed from ChatGPT on the web for Plus and Pro on September 14, 2026. Business workspaces and Enterprise admins have their own automatic-routing controls. For a deeper walkthrough of the modes, read our ChatGPT reasoning modes guide.

ChatGPT Plan Tiers and Context Windows in 2026

Your plan decides both which models you reach and how much of a conversation ChatGPT can hold: 27K tokens on Free, 54K on Go and Plus, and 128K on Pro for instant replies, rising to 256K on Go and Plus and 400K on Pro for reasoning. These in-app windows are far smaller than the 1,050,000-token API window of the same models, because ChatGPT shares the window with system instructions, memory and reasoning.

Two plan changes matter this month. Since September 29, 2026, Pro comes in three plans, Pro 100 ($100), Pro 200 ($200, open to new subscribers again with a lower allowance unless grandfathered) and Pro 500 ($500, the only plan with Astra Ultrafast). And OpenAI publishes no fixed message counts for Sol thinking levels on individual plans, only the GPT-6 Pro allowances listed above. For prices in detail see our ChatGPT pricing guide, and for usage caps see ChatGPT limits by plan.

How do I switch between ChatGPT models?

On paid plans, open the model picker in the message composer on the web, or at the top of the chat on mobile, and choose Instant, Medium, High, Extra High or Pro. On eligible plans the Pro option lists GPT-6 Pro and GPT-5.6 Sol Pro; OpenAI's instruction for Sol Pro is to select GPT-5.6 Sol in the model menu, then choose Pro. Free and Go users have no model choice, only the Think button.

Tools Across ChatGPT Models

Model choice now also decides which ChatGPT surface you use: Chat for conversation, ChatGPT Work for longer agent tasks and finished documents, and Codex for software work, each with its own model controls. ChatGPT Work launched on July 9, 2026, and it is where Plus users reach GPT-6 Astra and where paid users can pick Terra or Luna. On September 22, 2026 OpenAI added two more models to that surface, GPT-6 Sol and GPT-6 Luna, and says plainly that they "are not available in regular ChatGPT conversations", so the model list in Work and Codex is now longer than the one in Chat.

• Web search, file uploads, image creation and deep research: available across plans with plan-based limits
• Voice: since September 9, 2026, ChatGPT Voice can use GPT-5.6 or GPT-6 Astra when it needs to search or reason, with the same model controls as text chat
• Canvas: no longer available in GPT-5.5 Instant and Thinking since May 28, 2026; writing and code blocks appear directly in the chat
• Projects, scheduled tasks and custom GPTs: available with plan-based limits (personal plans can use existing custom GPTs but cannot create new ones); on September 11, 2026 OpenAI announced plans to retire custom GPTs and migrate them to plugins, with retirement scheduled for December 11, 2026
• Personalization: base styles such as Default, Friendly, Efficient, Professional, Candid and Quirky, with controls for warmth and emoji use; the Nerdy style was sunset in March 2026

Whatever the model, the output lives in the same chat history. AI Toolbox saves any single chat as TXT for free, and bulk export plus Markdown, JSON or PDF come with Premium.

Running Instant drafts, High-effort analysis and GPT-6 Pro threads side by side? AI Toolbox (formerly ChatGPT Toolbox) adds folders, full-text search and export to ChatGPT on every plan, including Free, so parallel threads stay findable no matter which model wrote them. 40,000+ users, 4.7/5 on the Chrome Web Store.

Install AI Toolbox free from the Chrome Web Store

Real-World Use Cases by Profession

The right ChatGPT model depends on the job as much as the plan, and for most professionals the answer is Instant for routine work and a higher thinking level only for the hard part.

Software developers

Use Instant for quick syntax questions and High for debugging and refactors. For long multi-file work, GPT-6 Astra in Codex is the option OpenAI positions for software engineering. On Pro, Extra High is the last step before GPT-6 Pro. See AI Toolbox for developers for keeping code threads organized.

Researchers and analysts

Use High for literature synthesis and data work, and ChatGPT Work with GPT-6 Astra when you want a cited report or spreadsheet as the output. Keep GPT-6 Pro for the questions where a wrong answer is expensive. See AI Toolbox for researchers.

Marketers and content writers

Instant handles drafts, emails and social posts; Medium helps when a long piece must follow a detailed brief. Decks and campaign reports built from templates are where Astra in Work earns its place. See AI Toolbox for marketers.

Students

Free and Go give unlimited text chats with GPT-5.6 Luna, enough for study guides and essay feedback. For math and science problems, press Think; on Plus, choose High. See AI Toolbox for students.

Business professionals

On Business seats, use Instant and Medium for meetings, proposals and email, and treat the shared Pro allowance (15 messages a month on Standard seats) as a budget for decisions that need it. See AI Toolbox for business professionals.

Tips for Getting Better Answers From Any ChatGPT Model

A clear prompt on Instant often beats a vague prompt on Extra High, so fix the prompt before you raise the thinking level.

• Be specific: state the goal, the audience, the format and any constraints
• Give context up front: paste the relevant document or code at the start, within your plan's context window
• Raise the level only when needed: Instant, then Medium or High; Plus and Pro no longer do this for you
• Use Work for deliverables: documents, spreadsheets and decks are what ChatGPT Work is built for
• Save prompts that work: reuse them with AI Toolbox's prompt library and the // shortcut
• Keep instructions consistent: set tone once in custom instructions instead of repeating it in every chat

How AI Toolbox Organizes ChatGPT Conversations Across All Models

ChatGPT's sidebar lists chats by date and does not group them by model, project folder or topic, so a week of Instant drafts, High-effort analysis and GPT-6 Pro threads ends up in one long list. AI Toolbox (formerly ChatGPT Toolbox), a Chrome extension with 40,000+ users and a 4.7/5 Chrome Web Store rating, adds the organization layer on top of ChatGPT. It works at the interface level, so it is unaffected by which model answered.

Folders are filled by searching for a chat and adding it, see the ChatGPT folders feature, and prompt chaining runs saved prompts in sequence. For teams, the Teams plan is $15 per seat per month, or $12 per seat per month billed annually, with a 2-seat minimum and a 14-day free trial, and it covers all four modules. One install also covers Gemini, Claude and Grok. Compare plans on the AI Toolbox pricing page.

"The ability to organize chats into folders and subfolders has been a lifesaver for my projects, allowing me to quickly locate specific information without having to scroll through endless conversations. ... The advanced search functionality is also incredibly powerful, allowing me to find specific messages in seconds."

Siddharth Katiyar, Chrome Web Store review

Limitations and What Native ChatGPT Handles Well

You do not need any extension to choose or use a ChatGPT model: the picker, thinking levels, Pro allowances and plan limits are all controlled by OpenAI, and nothing in AI Toolbox changes which model answers or how many messages you get. Native ChatGPT also does several things well on its own: Projects group related chats with shared files and instructions, the built-in search finds recent chats, and account export delivers your full history.

This guide has limits too. OpenAI changes model availability often, sometimes several times a month, and rollouts reach accounts at different times, so your picker can differ from the tables above for a few days. OpenAI does not publish fixed message counts for Sol thinking levels on individual plans, which legacy models remain for Plus and Pro, or confirmed Business and Enterprise context windows in its individual plan comparison. Benchmark scores quoted here are OpenAI's own, not independent tests. When in doubt, the model picker in your account is the source of truth.

Frequently Asked Questions

What is the latest version of ChatGPT?

The latest ChatGPT model is GPT-6 Astra, introduced by OpenAI on September 3, 2026. In regular chats it appears as GPT-6 Pro on the Pro, Business and Enterprise plans, and Plus users get Astra in ChatGPT Work and Codex. Everyday chats run GPT-5.6 Sol on paid plans and GPT-5.6 Luna on Free and Go.

Which ChatGPT model is best?

GPT-6 Astra is the most capable, but the best choice depends on the task. Use Instant (GPT-5.6 Sol) for quick answers and writing, High for coding, math and data analysis, and GPT-6 Pro for the hardest problems on Pro or Business. Free and Go users get GPT-5.6 Luna and can press Think for harder questions.

What ChatGPT versions are available in 2026?

In September 2026, ChatGPT offers GPT-6 Astra (as GPT-6 Pro and in Work and Codex), GPT-6 Sol and GPT-6 Luna in Work and Codex, GPT-5.6 Sol and GPT-5.6 Sol Pro on paid plans, GPT-5.6 Luna on Free and Go, and GPT-5.6 Terra in Work and Codex. GPT-4o, GPT-4.1, o4-mini, GPT-5.1, GPT-5.2 and GPT-4.5 have been retired from ChatGPT.

Can I use GPT-6 Sol or GPT-6 Luna in a normal ChatGPT chat?

No. OpenAI introduced GPT-6 Sol and GPT-6 Luna on September 22, 2026 for ChatGPT Work and Codex, and its help centre states that they "are not available in regular ChatGPT conversations". Availability depends on your plan, workspace settings and rollout access, and Work and Codex allowances are counted separately from Chat.

Which ChatGPT model is the most advanced?

GPT-6 Astra is the most advanced model in ChatGPT. OpenAI reports it scores 57.9% on Terminal-Bench 4.0 against 37.3% for GPT-5.6 Sol, and it has a 1,050,000-token context window in the API. In Chat it is available as GPT-6 Pro on Pro $100, Pro $200, Business and Enterprise plans, with usage limits that rise with the plan's allowance.

Is Instant or Thinking better in ChatGPT?

Neither is better for everything. Instant is faster and fine for everyday questions and writing. Medium, High and Extra High, which replaced Thinking Standard, Extended and Heavy in June 2026, spend more time reasoning and are better for coding, math and multi-step analysis. Since September 14, 2026, Plus and Pro chats no longer switch from Instant automatically.

What is the difference between GPT-5.6 Luna and GPT-5.6 Sol?

Luna is the fastest and lowest-cost GPT-5.6 model and the default on Free and Go, with unlimited text chats. Sol is the GPT-5.6 flagship that powers Instant, Medium, High and Extra High on paid plans. Free and Go users cannot use Sol, and their Think button runs on Luna.

Can free ChatGPT users use GPT-6 Astra?

No. GPT-6 Astra is not available on Free or Go. Free and Go accounts use GPT-5.6 Luna with unlimited everyday text chats and a Think button for harder questions. Plus includes Astra in ChatGPT Work and Codex, while GPT-6 Pro in regular chats requires Pro, Business or Enterprise.

Can I still use GPT-4o or GPT-4.5 in ChatGPT?

No. OpenAI retired GPT-4o from ChatGPT on February 13, 2026, with custom GPTs on Business, Enterprise and Edu keeping it until April 3, 2026. GPT-4.5 was retired from ChatGPT on June 26, 2026. GPT-4 left on April 30, 2025. Some older models remain available in the OpenAI API.

Key Takeaways

• Latest model: GPT-6 Astra, introduced September 3, 2026; GPT-6 Pro in Chat on Pro, Business and Enterprise, and in Work and Codex on Plus.
• Everyday chats: GPT-5.6 Sol on paid plans, from Instant to Extra High; GPT-5.6 Luna on Free and Go with unlimited text chats.
• Best pick: Instant for routine work, High for hard analysis, GPT-6 Pro for the hardest problems within its allowance.
• In-app context: 27K on Free, 54K on Go and Plus, 128K on Pro for instant replies; 256K and 400K for reasoning (OpenAI pricing page).
• No more auto-switching: since September 14, 2026, Plus and Pro pick thinking levels manually.
• Retired in 2026: GPT-4o, GPT-4.1, o4-mini and GPT-5 (February 13), GPT-5.1 (March 11), GPT-5.2 (June 12), GPT-4.5 (June 26).

Bottom Line

ChatGPT in September 2026 runs GPT-6 Astra at the top and the GPT-5.6 family underneath. Most paid chats are answered by GPT-5.6 Sol at the thinking level you choose, Free and Go run on GPT-5.6 Luna, and GPT-6 Astra is reserved for GPT-6 Pro, ChatGPT Work and Codex. Pick the lowest level that gets the job done and spend Pro allowances on the problems that deserve them. To keep conversations from every model organized and searchable, install AI Toolbox or read how to organize ChatGPT conversations.

Sources

• OpenAI Help Center, GPT-5.6 and GPT-6 Pro in ChatGPT (plans, thinking levels, GPT-6 Pro allowances)
• OpenAI Help Center, ChatGPT Release Notes (Astra launch, auto-switching change, retirements, picker rename)
• OpenAI, ChatGPT pricing (plan models and context windows)
• OpenAI, GPT-6 Astra: A new generation of intelligence
• OpenAI, GPT-5.6: Frontier intelligence that scales with your ambition (July 9, 2026)
• OpenAI API documentation, GPT-6 Astra and the GPT-5.6 Sol, Terra and Luna model pages
• OpenAI Help Center, Retiring GPT-4o and other ChatGPT models
• OpenAI Help Center, About ChatGPT Pro tiers (Pro 100, Pro 200 and Pro 500, checked October 5, 2026)
• OpenAI Help Center, ChatGPT Work and Codex (GPT-6 Sol and GPT-6 Luna, September 22, 2026; Astra allowances in Work and Codex)

Last updated: October 5, 2026