ChatGPT Free vs Go vs Plus After GPT-6: Which Model You Get

Free and Go both run GPT-6 Luna in ChatGPT's Chat tab. Plus ($20 a month) is the cheapest plan that gets GPT-6 Sol, and the only one under $100 with Codex outside the desktop app. Go ($8) buys more room, not a better model: double the Instant context window, more uploads and voice, and scheduled tasks. If you write code with ChatGPT, the upgrade that matters is Free to Plus, and the reason is Codex, not chat quality.

Last updated: October 9, 2026. Every price, model and limit below was checked that day against OpenAI's announcement, pricing pages and help center.

Why now: on October 7 OpenAI moved Chat to GPT-6 with a new answer format called Intelligent UI, rolling out to Plus, Pro, Business and Enterprise that day and to Free and Go from October 8. The same post splits the family by plan: "GPT-6 Sol for Plus, Pro, Business, and Enterprise tiers, and GPT-6 Luna for Free and Go tiers." That turns "is Plus worth it" into a model question again. The Hacker News thread on the launch (747 points and 453 comments when we checked on October 9) asked it directly.

Official Sources#

What Each Plan Gets Now#

All rows come from the ChatGPT pricing page and the models help article, except the Codex row, which comes from the Codex pricing page, and the Free ads note, which comes from OpenAI's October 5 ads announcement.

Two things in that table are easy to miss. First, Go and Plus have the same Instant context window (54K) and the same reasoning window (256K), so the $12 gap between them is not about how much text ChatGPT can hold. It is the model (Luna vs Sol) and Codex. Second, Plus does not get Extra High in Chat. The help center's availability table lists Medium and High as included on Plus and Extra High as "Not included." If you want GPT-6 Sol at its top thinking level in Chat, that is a Pro feature.

The Docs Disagree About Free and Go#

As of October 9, OpenAI's own pages do not agree on what Free and Go run. The announcement and the pricing page's model table both say GPT-6 Luna. The models help article still says Instant on Free and Go uses GPT-5.6 Luna, and that the Think button "uses GPT-5.6 Luna, not GPT-5.6 Sol."

Our read: the help article predates the Free and Go rollout that started October 8 and has not caught up. Two practical consequences. If you are on Free or Go and your answers look like the old format, you may simply not have the rollout yet. And the more useful fact is the one both pages agree on: neither Free nor Go gets any version of Sol in Chat. Whatever the Luna version number, the step up to Sol starts at $20.

Go vs Plus: Where the $12 Goes#

Go is the plan the GPT-6 launch squeezed. It already shared a model with Free under GPT-5.6, when Luna became the free default, and GPT-6 keeps it that way. What $8 buys is capacity:

• Context: 54K Instant context instead of 27K, and ~40 pages of input instead of ~12. For pasting a long log or a spec into chat, that is the single biggest difference between Free and Go.
• Limits: more messages with tools, more uploads, more image creation, more voice, longer memory.
• Features: scheduled tasks, voice with video, and creating and sharing plugins, none of which Free gets.

What it does not buy is a smarter model, Codex beyond the desktop app, or the "Fast" response tier. Plus adds GPT-6 Sol in Chat, expanded memory and deep research, Sites, developer mode, interactive tables and charts, and real Codex access. If you would pay for Go to get better answers, you will not get them: pay for Plus or stay on Free.

Why Developers Should Judge Plus by Codex, Not Chat#

OpenAI says plainly that "the models powering Work and Codex are not changing as part of this release." The GPT-6 launch was a Chat launch. For coding, the plan line that matters is the Codex row, and it is not close.

On Free and Go, the Codex pricing page lists one thing: "GPT-6 Luna at Standard speed in the desktop app, subject to rollout." No CLI, no IDE extension, no cloud tasks. On Plus you get Codex on the web, in the CLI, in the IDE extension and on iOS, plus cloud integrations like automatic code review.

Which models Plus gets in Codex depends on which OpenAI page you read. The Plus plan card on the Codex pricing page lists only "GPT-6.1 Sol and GPT-6 Luna." The usage table further down the same page gives Plus estimates for four models, adding GPT-6 Sol and GPT-6 Astra, and the models help article says "Plus plans include GPT-6 Astra in ChatGPT Work and Codex." The ChatGPT pricing page's model table also marks GPT-6 Sol and GPT-6 Astra as included on Plus, without saying on which surface. The plan card is the narrowest of the four, so treat GPT-6.1 Sol and GPT-6 Luna as the guaranteed pair, and check the Codex model picker for Sol and Astra. OpenAI's estimates for local Codex messages per five-hour window on Plus:

These are OpenAI's ranges, not fixed limits, and weekly caps may also apply. Two notes for Plus buyers. GPT-6 Astra is on Plus only in Work and Codex, not in Chat; GPT-6 Pro in Chat starts at Pro 100. And GPT-6.1 Sol is also Work and Codex only. Our GPT-6.1 Sol guide covers what it changes, and the Codex usage limits breakdown covers how the allowance drains.

Housekeeping: GPT-5.5 retires from ChatGPT, Work and Codex on all plans on October 14, 2026, per the same pricing page. If a saved Codex profile or script pins it, move it to a GPT-6 model this week.

Or Skip the Subscription: API Prices#

If what you actually want is a cheap model for scripts, the API may beat every plan. From OpenAI's model pages, per million tokens:

A worked example: 1 million input tokens and 200,000 output tokens a month costs $0.20 on Luna ($0.10 input plus $0.10 output) and $4.00 on Sol ($2.00 plus $2.00), as long as each request stays under 272K input tokens. Both models take a 1,050,000-token context window through the API, far more than any Chat plan exposes, but using it costs more. Per both model pages, "Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request." The surcharge applies to the whole request, not just the tokens past the line, so a 300K-token Luna prompt bills every input token at $0.20 per million and its output at $0.75 per million. If you plan to paste whole repos or long transcripts, budget for the higher rate or trim below 272K. The other catches: the API has no free tier for these models (the rate-limit table lists Free as "Not supported"), you do not get Intelligent UI, and API spend does not come with a Codex allowance. Codex will run on an API key, billed at API rates, but you lose cloud features like GitHub code review. Our GPT-6 family explainer has the full model-by-model breakdown.

Who Wins, Who Loses#

Free users win the most. A GPT-6-generation model with unlimited everyday text chat for $0 resets the baseline every rival free tier gets compared against.

Go loses its pitch. When Free and Go share a model, Go becomes a capacity plan, and the pricing card's note that it "may include ads" makes the trade sharper: $8 for more room, possibly with ads, and the same answers. Paying for Go does not opt you out of ads either: OpenAI's ads announcement says it will "initially" test its new visual ad format "during image generation in ChatGPT for users on the Free and Go plans."

Plus holds, but for a different reason than a year ago. Its edge over Free used to be a better chat model; now the bigger gap is agentic work: Codex across every surface, GPT-6.1 Sol and, per the help center, a slice of Astra. The second-order effect is where OpenAI's upsell goes next. Once the free chat is good enough for most people, paid plans have to sell agent allowances, not chat quality. That is already visible in the plan cards: the Pro tiers are differentiated by Codex and Work allowances and by Ultrafast, which only Pro 500 gets, per the Pro tiers article. Expect the next round of plan changes to be about metering agents, not about which chat model you get. For the cross-vendor version of this decision, see ChatGPT Plus vs Claude Pro for developers.

What People Are Actually Saying#

• The worth-paying question is the top one. In the Hacker News thread, user rc-1140 asked "why is this worth paying for as a layman?", adding that they can sometimes find a use for LLMs in software development but never "outside of 'slightly better search engine'". That matches the table above: for chat alone, Free now covers most people.
• GPT-6's design taste splits the room. User kingstnap wrote that "GPT-6's design sense is kind of ridiculous" and that they keep "explicit instructions to tone it down," adding "Hopefully this doesn't bleed into the chat..." The same argument broke out over OpenAI's launch page itself: lrpe called pages like it "horrible" and ended with "Just give me text to read," and mikestorrent replied that "Sometimes you want more than text," pointing to "a very legitimate need to show tables, graphics, interesting graphs." If you fall in the text camp in Chat, the help article documents a Layout and visuals toggle in web Settings that reduces visual elements, though it is not a full text-only mode.
• Ads suspicion, with a counterpoint. User virtuosarmo wrote that the new UI "seems like a way for them to surface ads in ChatGPT," noting that OpenAI had launched a visual ad format option for advertisers a couple of days earlier (the October 5 announcement). Further down that sub-thread, aroman noted that "ChatGPT has had ads in the product for the better part of a year now." The plan-level facts on record: the Go pricing card's ads note, and OpenAI's statement that the visual ad format test starts on Free and Go during image generation. Neither source says Plus or Pro show ads.
• Capacity is not unlimited even at the top. User wincy reported hitting "model not available" limits for GPT-6.1 Sol "multiple times over the last week," on both a personal $200 Pro plan and a Codex enterprise connection. "It's never for very long," they added, and they could switch to Astra. One person's experience, but it is a reminder that a higher tier buys allowance, not guaranteed availability.

Which Plan to Pick#

FAQ#

What model does ChatGPT Free use now?#

OpenAI's announcement and pricing page say Free and Go run GPT-6 Luna in Chat, rolling out from October 8, 2026. The help center article still says GPT-5.6 Luna as of October 9, so you may see the older model until the rollout reaches your account. Neither plan gets GPT-6 Sol.

Is ChatGPT Go worth it over Free?#

Only if you need capacity. Go ($8 a month) gets the same Chat model as Free but doubles the Instant context window to 54K, raises the input maximum from about 12 pages to about 40, and adds scheduled tasks and more uploads, images and voice. OpenAI notes that Go may include ads.

Is ChatGPT Plus worth it after GPT-6?#

For developers, yes, mainly because of Codex: Plus adds the CLI, IDE extension and web. The Codex plan card lists GPT-6.1 Sol and GPT-6 Luna; OpenAI's usage table and help center also list GPT-6 Sol and limited GPT-6 Astra on Plus. For chat alone, Plus buys GPT-6 Sol at Instant, Medium and High, which matters less now that Free gets a GPT-6 model.

Does ChatGPT Plus include GPT-6 Astra?#

In Work and Codex, yes, according to the help center, with limited usage (OpenAI estimates 5-45 local Codex messages per five hours). The Plus card on the Codex pricing page does not list Astra, so check the model picker on your account. In Chat, no: GPT-6 Pro, powered by Astra, requires Pro 100, Pro 200, Business or Enterprise.

Can I use GPT-6 Luna without a subscription?#

Yes, through the OpenAI API as gpt-6-luna at $0.10 per million input tokens and $0.50 per million output tokens. The API has no free tier for it, requests over 272K input tokens are billed at 2x input and 1.5x output, and you do not get Intelligent UI or a Codex allowance.

Sources#

Continue Reading#

• GPT-6 in 7 Minutes: Astra, Sol, and Luna Explained for Developers - the full GPT-6 family, API prices and new API features
• OpenAI Retunes GPT-5.6 Sol in ChatGPT and Makes Luna the Free Tier Default - the August change that first put Luna on Free and Go
• GPT-6.1 Sol Release Guide: Near-Astra Agentic Work at $2/$10 and $0.10 Cache Reads - the Work and Codex model Plus gets that Chat does not
• Codex Usage Limits and Pricing in 2026 - how the Plus Codex allowance drains and what credits cost
• ChatGPT Plus vs Claude Pro for Developers - the same $20 decision across vendors