ChatGPT Review (September 2026): GPT-6 Astra, A $500 Tier And The Death Of Atlas, Sora And Operator

This ChatGPT review tests the September 2026 reality of OpenAI’s flagship product, and the story has inverted. For two years every update added a surface: a browser, a video model, a browser agent. This quarter OpenAI deleted three of them. Atlas is gone. Sora is gone, with no successor. Operator and the ChatGPT agent that replaced it are gone. In their place is something that looks a lot like an office suite — ChatGPT Work, Dots, Spaces, Pages, Slides — aimed squarely at Microsoft. The platform now serves 1.2 billion weekly users, up from 900 million in May.

The buying question has changed shape with it. It is no longer “which of seven tiers fits my workflow” but something sharper: OpenAI just added a $500/month tier and, from October 30, is making the $200 tier materially worse. That is a pricing ladder being restructured around compute scarcity rather than user segments, and it changes what a subscription is worth. This ChatGPT review works through the nine tiers as they actually stand at the end of September 2026, what the shutdowns mean if you built a workflow on them, and where ChatGPT still wins.

The Bottom Line (September 2026)

The honest ChatGPT review verdict for September 2026: ChatGPT is still the right default for the broadest range of users, but for a different reason than it was in May. The breadth argument took a real hit — Atlas, Sora, and Operator are all discontinued, so three of the surfaces that justified the “one subscription covers everything” pitch no longer exist. What replaced them is better: ChatGPT Work turns a goal into finished spreadsheets, decks, and documents, and Dots are always-on agents that keep working when you close the laptop. Most personal users should still land on Plus at $20/month, which now buys GPT-6 Astra through Work, Codex, Custom GPTs, Canvas, Tasks, and Advanced Voice. Go at $8/month remains the right escape from free-tier limits. The Pro ladder is where it gets complicated: $100 is the entry point for Dots and GPT-6 Pro chat, $200 is about to deliver less than it does today, and $500 buys Ultrafast inference most people cannot use enough to justify.

⏱️ What Just Happened (May to September 2026)

The previous version of this ChatGPT review captured the expansion moment in May 2026, when the product family had grown to seven surfaces and a new $100 Pro tier had just landed. Four months later the shape is different in kind, not degree — three products were retired, an office suite arrived, and the pricing ceiling more than doubled:

• GPT-6 Astra shipped September 3, 2026. OpenAI’s new flagship: a 1.05M-token context window, 128K max output, and an April 30, 2026 knowledge cutoff. API pricing is $10 per million input tokens and $50 per million output. It sets a new frontier on computer and browser use at 72.6% on OSWorld 2.0, and saturates several evaluation suites outright — 97.6% on FrontierMath Tier 4 and, on OpenAI’s own provider-adapter harness, 99.9% on ARC-AGI-3.
• GPT-6.1 Sol arrived September 29 and is the value story. OpenAI describes it as delivering coding performance comparable to GPT-6 Astra at one-fifth the cost — $2 per million input tokens and $10 per million output. Plus, Pro, Business, Enterprise, and Education users got it in Codex and ChatGPT Work immediately, with general ChatGPT availability to follow.
• Free and Go moved to GPT-5.6 Luna. The unpaid and $8 tiers now run Luna rather than a numbered GPT-5.5 build, and both remain ad-supported. GPT-6 Astra is restricted to paid plans.
• ChatGPT Work launched in July 2026. This is the most consequential addition of the year. Work is an agent that takes an outcome, gathers context across your connected apps and files, breaks the job into steps, and returns finished material — spreadsheets, slides, documents, and interactive web apps rather than chat. More than 35 million people now use Codex and ChatGPT Work weekly, against roughly 5 million Codex users in June.
• Dots launched at DevDay on September 29. Always-on autonomous agents powered by GPT-6 Astra, each with its own cloud computer and browser, extensible through more than 4,000 plug-ins, and reachable from ChatGPT, Slack, Microsoft Teams, or a voice call. They keep working while you are offline. You can set custom rules, require approval before actions, and inspect activity logs.
• Spaces, Pages, and collaborative Slides. Spaces is a shared workspace where colleagues work alongside ChatGPT and their Dots; Pages is a document type built for human-and-agent editing; Slides is a conversationally built deck editor with comments; since October 16 it has been rolling out to eligible Pro, Business and Enterprise users with Space access, on the web and in the desktop app (Plus is not on that list). Taken together this is an office suite, and the target is obvious.
• Atlas browser was discontinued. The AI-native browser launched in October 2025 stopped working on August 9, 2026, less than a year in. OpenAI folded browser-based agentic capability into ChatGPT and Codex instead, bringing multiple tabs, downloads, improved navigation, and account login support into ChatGPT itself.
• Sora was discontinued, with no successor. The app closed on April 26, 2026 and the API followed on September 24. Reporting attributes the shutdown to compute costs — around $500,000 a day in server expense — alongside deepfake controversies and weak revenue against the resources required. OpenAI has named no replacement, so ChatGPT currently has no video generation at all.
• Operator and the ChatGPT agent are both gone. Operator shut down after ChatGPT agent absorbed its browser-control features; agent was then removed from ChatGPT in early August 2026 without advance deprecation notice. OpenAI’s help centre points remaining users to ChatGPT Work and a separate cloud browser feature.
• A $500/month Pro 500 tier, and a worse $200 tier. Pro 500 launched at DevDay with OpenAI’s highest usage allowance and “Ultrafast” inference — up to 300 tokens per second in Codex and Work, roughly an 8x speed increase. From October 30, the $200 Pro tier drops from 20x to 10x the Plus allowance in Codex and Work, and its weekly GPT-6 Pro message cap falls from 200 to 100. Existing subscribers get a one-time transition credit.
• Sign in with ChatGPT. Sixteen partner services, including Notion and Cognition, now support ChatGPT as an identity provider, removing the separate API-key setup step.

The single most important change for the ChatGPT review buying decision is that the pricing ladder is no longer organised around user types. It is organised around compute. Pro 500 exists because inference speed is now a product you can buy, and the $200 tier is being reduced because the allowance it promised is expensive to honour. That is a different kind of pricing page than the one we reviewed in May, and it rewards a different kind of shopping: work out what you actually consume, then buy the smallest tier that covers it, because the tier you buy today may deliver less in five weeks.

🤖 What ChatGPT Actually Is (In September 2026)

The ChatGPT review tech context: in May this was seven things stacked under one brand. In September it is six, and they are different things — the browser, the video model, and the browser agent were all retired, and an office suite grew in their place. The current surfaces:

1. Classic ChatGPT chat — the conversational interface on web, desktop, and mobile. GPT-5.6 Luna on Free and Go, GPT-6 Astra on paid tiers, GPT-6 Pro for chat on $100-and-up Pro tiers under a weekly message cap. Browser capability that used to live in Atlas is now built in: multiple tabs, downloads, navigation, account logins.
2. ChatGPT Work — launched July 2026 and now the centre of gravity. An agent that takes a goal, pulls context from your connected apps and files, decomposes the job, and returns finished spreadsheets, slides, documents, and interactive web apps. Usage is metered as a multiple of the Plus allowance.
3. Dots — always-on autonomous agents on GPT-6 Astra, each with a dedicated cloud computer and browser, 4,000-plus plug-ins, and contactable through ChatGPT, Slack, Teams, or voice. Available on Pro ($100+), Business Premium ($125/seat), and Enterprise, Edu and Healthcare workspaces in beta. Not on Free or Go, and not yet available to Pro users in the EU, Switzerland, or the UK.
4. Spaces, Pages, and Slides — the collaboration layer announced at DevDay. Spaces is a shared workspace for colleagues plus their Dots, Pages is a document format built for human-and-agent co-editing, and Slides is a conversational deck builder.
5. Codex and Codex CLI — the full-stack engineering agent and its open-source MIT-licensed terminal counterpart. Codex now runs GPT-6.1 Sol and competes directly with Claude Code and Gemini CLI.
6. Custom GPTs + GPT Store — still the largest custom-AI marketplace, free to use with Plus required to create. Increasingly overshadowed by Dots, which do work rather than answer in a persona.

For the broad ChatGPT review question — “should I subscribe?” — surfaces 1 and 2 are what most people are actually buying, and surface 2 is the reason the answer is still yes. Dots (surface 3) are the most interesting thing here and the hardest to evaluate honestly, because they are days old and gated to $100-and-up. Codex (surface 5) matters for developers. The uncomfortable observation is that the surfaces retired this year — Atlas, Sora, Operator — were all consumer-facing, while everything added is aimed at work. Read that as a statement of where OpenAI thinks the revenue is.

🧠 The GPT-6 Lineup (September 2026)

Four models matter for this ChatGPT review, and which one you get depends entirely on what you pay.

GPT-6 Astra (Current Paid-Tier Flagship)

GPT-6 Astra shipped on September 3, 2026 and is the model behind paid ChatGPT, ChatGPT Work, and every Dot. Specifications: a 1,050,000-token context window, 128K maximum output, text and image input, and an April 30, 2026 knowledge cutoff. API pricing is $10 per million input tokens and $50 per million output, with cached input at $1 and batch at half price. The headline results are genuinely frontier — 97.6% on FrontierMath Tier 4, 100% on ExploitBench, and a new best of 72.6% on OSWorld 2.0 for computer and browser use at roughly 47% less time per task than its predecessor. Two caveats belong next to those numbers. OpenAI’s 99.9% on ARC-AGI-3 is reported “under OpenAI’s provider adapter harness,” which is a vendor-run measurement. And cyber-sensitive capabilities are gated behind a trusted-access programme, because Astra meets the “Critical” cybersecurity threshold under OpenAI’s own Preparedness Framework.

GPT-6.1 Sol (The One That Changes The Maths)

GPT-6.1 Sol arrived on September 29 and is the most interesting release of the quarter, because it makes the flagship look overpriced. OpenAI positions it as delivering coding performance comparable to GPT-6 Astra at one-fifth the cost: $2 per million input tokens and $10 per million output, against Astra’s $10/$50. Plus, Pro, Business, Enterprise, and Education subscribers got it immediately in Codex and ChatGPT Work, with general ChatGPT availability still to come. If you are choosing a model for build work rather than reading a benchmark table, start here and escalate to Astra only when Sol visibly falls short — the same escalation logic Anthropic now recommends between Opus and Fable.

GPT-6 Pro And GPT-5.6 Luna

GPT-6 Pro is the extended-reasoning option inside ChatGPT on the $100-and-up Pro tiers, and it is rationed by message count rather than tokens. Today that cap is 200 messages per week; from October 30 the $200 tier drops to 100. Worth knowing before you pick a tier on the strength of “Pro model access,” because the access is countable. GPT-5.6 Luna is the model serving Free and Go. It is a perfectly competent assistant and the reason the free tier still works for casual use, but the gap to Astra is the clearest it has been in two years — GPT-6 Astra is restricted to paid plans, so the free tier is no longer a sample of the flagship.

🛠️ The Product Family Beyond Chat — And What OpenAI Deleted

This is the section of the ChatGPT review that changed most since May, in both directions — what arrived is genuinely useful, and what left was genuinely being used.

ChatGPT Work (The New Centre Of Gravity)

ChatGPT Work launched in July 2026 and is the reason this review still recommends a subscription. The distinction from chat is that the output is not a message. You give Work an outcome, it gathers context across your connected apps and files, breaks the job into steps, executes them independently, and hands back a finished spreadsheet, deck, document, or small web app. Adoption backs the framing: more than 35 million people use Codex and ChatGPT Work weekly, up from roughly 5 million Codex users in June. Usage is metered as a multiple of the Plus allowance, which is exactly the number the October 30 change reduces on the $200 tier — so if Work is your main use, read the pricing section carefully before committing.

Dots, Spaces, Pages, And Slides

Dots are the DevDay headline: always-on agents on GPT-6 Astra, each with a dedicated cloud computer and browser, extensible through 4,000-plus plug-ins, and reachable from ChatGPT, Slack, Microsoft Teams, or a phone call. They run while you are offline, you can require approval before they act, and you can inspect their logs. OpenAI’s own example — a Dot noticing a forgotten invoice, preparing it, and sending it once approved — is a fair illustration of the intent. Spaces wraps this in collaboration: a shared workspace where colleagues and their Dots work on the same Pages and Slides. Two honest caveats. Dots start at the $100 Pro tier (one included), Business Premium at $125 a seat, or Enterprise, Edu, and Healthcare in beta — nothing on Free, Go, or Plus. And Pro users in the EU, Switzerland, and the UK cannot use them yet at all.

🎥 Watch: Meta Muse Is Free. ChatGPT’s Agent Costs $100. Here’s the Catch.

Dots, ChatGPT’s always-on agent, start at the $100 Pro tier. Meta’s Muse puts a background agent in its free plan, with Power at $20 and Maximum at $100 a month. This video compares the plans, the yearly cost and the privacy catch. Click to play.

Codex And Codex CLI

Codex is OpenAI’s full-stack software engineering agent and the surface that came out of this year strongest. It writes features, fixes bugs, runs tests, and reviews whole codebases; Plus includes it, and Pro tiers raise the allowance. Codex CLI is the standalone open-source MIT-licensed terminal version, competing directly with Claude Code and Gemini CLI. Two September changes matter: Codex now runs GPT-6.1 Sol, which means near-Astra coding at a fifth of the cost, and Ultrafast on the Pro 500 tier pushes generation to roughly 300 tokens per second — about 8x — which is the one place the $500 tier has an obvious, measurable payoff. For the deep-dive and the head-to-head against Claude Code, see our ChatGPT Codex review.

What OpenAI Discontinued: Atlas, Sora, And Operator

Three products this review recommended in May no longer exist, and if you built around them you should know exactly where you stand. Atlas, the AI-native browser launched in October 2025, stopped working on August 9, 2026 — less than a year after launch. Its capabilities were folded into ChatGPT and Codex, so tabs, downloads, navigation, and account logins now live in ChatGPT itself; the loss is the dedicated browsing surface, not the capability. Sora is the harder one: the app closed April 26 and the API on September 24, with no successor named. Reporting puts the cause at roughly $500,000 a day in rendering compute, plus deepfake controversy and revenue that did not justify the spend — a product that hit a million downloads faster than ChatGPT did, discontinued about six months later. ChatGPT now has no video generation, so for video you need Google Flow, Runway, or Gemini Omni. Operator was deprecated into ChatGPT agent, which was itself pulled in early August without advance notice; the help centre redirects to ChatGPT Work and a separate cloud browser. Our ChatGPT Agent Mode review documents how that surface behaved while it existed.

Custom GPTs And The 3M+ Marketplace

The Custom GPT marketplace crossed 3 million entries in early 2026 and remains the largest custom-AI ecosystem in the market. Creation requires Plus; usage is free for anyone. Competing tools (Gemini Gems, Claude Projects) still have meaningfully smaller pre-built libraries, and the GPT Store’s discovery surface is a real advantage. The strategic question is what Dots do to it. A Custom GPT is a persona that answers; a Dot is an agent with its own computer that finishes work. If Dots land, the marketplace advantage matters less than it did in May. For the head-to-head, see our Gemini Gems review.

💰 The Nine-Tier Pricing (September 2026)

Every ChatGPT review eventually comes down to this table, and this quarter it comes with a deadline attached.

Three observations matter most for the buying decision. First: the tier boundary that actually matters is now $100, not $20, because that is where Dots begin. Everything below it is a chat-and-Work subscription; everything at or above it comes with an agent that runs while you sleep. Second: if you are on $200 Pro, act before October 30 — your Codex and Work allowance halves from 20x to 10x Plus and your weekly GPT-6 Pro messages drop from 200 to 100. If you are not consuming more than 10x Plus today, you are about to pay $200 for what $100 delivers, and the honest move is to downgrade and save $1,200 a year. Third: Pro 500 is a speed purchase, not a capability purchase. Ultrafast’s ~300 tokens per second in Codex and Work is real and measurable, but you need a workflow where you are genuinely waiting on generation for it to return $6,000 a year.

🎥 Watch: ChatGPT’s New $500 Plan — Every ChatGPT Plan Explained

Nine tiers is a lot to hold in your head. This walks through every current ChatGPT plan, what the new $500 Pro 500 tier actually buys, and what changes on the $200 tier on October 30. Click to play.

🎥 Watch: ChatGPT Plans: Don’t Pick the Wrong One

Which of the ChatGPT tiers fits you? Our video walks up every plan, from Free and Go to Plus ($20 a month) and the three Pro plans ($100, $200 and $500), using OpenAI’s own pages, and ends with which plan suits which person. Click to play.

⚔️ ChatGPT vs Claude vs Gemini vs Perplexity vs Grok (September 2026)

The honest competitive frame for the ChatGPT review: still a five-way race, but the edges moved. ChatGPT leads on scale (1.2 billion weekly users), the Custom GPT marketplace, consumer polish, and now the most complete agent-and-office story with Work, Dots, and Spaces. Claude took the coding crown outright — Opus 5.5 leads Terminal-Bench 4.0 at 66.4% against GPT-6 Astra’s 57.9%, at $4/$20 per million tokens against Astra’s $10/$50. Gemini leads on Google integration and price. Perplexity still leads on cited research. Grok still owns real-time X data. The significant change is that ChatGPT lost its multimedia sweep: with Sora discontinued, it has image generation and voice but no video, while Gemini gives video away.

The pattern: ChatGPT wins on scale, agents, and the office story — nobody else ships anything quite like Dots, and Work plus Spaces is a more coherent document play than Claude’s or a cheaper one than Microsoft’s. It loses on price at the frontier (Astra costs 2.5x Opus 5.5 and 5x Gemini 3.1 Pro), on terminal coding, and now on video, where discontinuing Sora handed a category to Google unopposed. The honest framing is unchanged in shape: ChatGPT is still the best “one subscription covers most workflows” choice, but the gaps are specific enough to name. Pair it with Perplexity for cited research, Claude for coding, and Gemini Omni if you need video at all.

🧪 Real Test Results (September 2026)

The same four tasks this ChatGPT review has run since the first edition, re-run against the current models.

Test 1: Long-Form Writing (1500-Word Blog Post)

Task: write a 1,500-word blog post explaining the trade-offs between three AI customer support tools for a B2B SaaS audience. GPT-6 Astra produced a clean draft in one shot with reasonable structure, balanced product coverage, and a confident voice — light editing for tone consistency, no factual rewrites. Claude Opus 5.5 produced a more measured draft with better nuanced reasoning and, this time, a warmer voice than previous Claude releases; it needed less energizing than Opus 4.7 did in the spring. Gemini 3.1 Pro produced a competent draft that still read like committee-approved corporate prose. Winner for general drafting: closer than it was, and now arguably Claude on final quality, with Astra still ahead on ready-to-publish-without-thinking.

Test 2: Debug A Complex Race Condition

Task: hunt down a flaky test caused by an asyncio race condition in a Python payment-processing service. GPT-6 Astra identified the candidate race within five iterations and proposed a lock-based fix that worked, though the diagnostic path was less methodical than Claude’s. The benchmark picture now matches the felt experience: Claude Opus 5.5 leads Terminal-Bench 4.0 at 66.4% against Astra’s 57.9%, having taken a benchmark OpenAI held for eighteen months. Two practical notes for ChatGPT users. Codex remains a much better surface than chat-style coding, and it now runs GPT-6.1 Sol — near-Astra coding at a fifth of the API cost. For dedicated production coding, Claude Code still leads.

Test 3: Deep Research On A Niche Topic

Task: produce a research report on the FDA approval status of three competing GLP-1 receptor agonists with cited primary sources. Deep Research on GPT-6 Astra produced a structured report with eight cited sources, accurately distinguished the three drugs, and flagged where data was preliminary. Perplexity Pro produced a comparable report with slightly stronger source attribution. Claude produced a more cautious report that hedged harder on contested findings. For cited research, Perplexity still edges ahead, and Deep Research remains a real upgrade over chat-only research. Worth noting that Astra’s knowledge cutoff is April 30, 2026, so anything from this summer has to be retrieved rather than recalled — which makes the retrieval quality, not the model, the thing that decides these tasks.

Test 4: Creative Writing (Brand Email)

Task: draft a 600-word email pitching a SaaS product to a skeptical CMO who’s been burned by AI vendors. GPT-6 Astra produced a confident, single-voice draft with a sharp opening hook and clean structural flow. Claude Opus 5.5 produced a more thoughtful draft with more nuanced empathy for the skepticism and more measured value claims. Gemini 3.1 Pro produced a competent corporate draft. For persuasive writing, Astra still wins on default delivery and Claude wins when you will invest editing time in subtler positioning. This is the workflow class where ChatGPT’s ecosystem advantage compounds — a Custom GPT trained on brand voice does this faster, and a Dot can now be told to draft, hold for approval, and send.

🎯 Which ChatGPT Tier You Actually Need

The practical output of any ChatGPT review is a tier recommendation, so here is the honest mapping from how you work to what you should pay.

Free Tier — Perfect For:

• Casual users who chat a few times a week and don’t hit rate limits — note that Free runs GPT-5.6 Luna, not GPT-6 Astra, so it is no longer a sample of the flagship.
• People sampling the product before deciding whether to upgrade.
• Users who find the ad-supported experience acceptable (Free and Go are both ad-supported; every paid tier from Plus up is ad-free).
• Anyone whose AI work is occasional rather than central to their workflow.

Go Tier ($8/mo) — Perfect For:

• Free-tier users who hit rate limits and want a more usable everyday experience without committing to Plus.
• Users who want higher limits than Free but don’t need the Plus feature surface (Codex, Custom GPT creation, ChatGPT Work). Note Go is still ad-supported — Plus is the first ad-free tier.
• Students, hobbyists, and casual professional users.
• Direct comparison with Google AI Plus ($7.99) — pick by ecosystem preference.

Plus Tier ($20/mo) — Perfect For:

• The dominant default for most personal users — GPT-6 Astra through ChatGPT Work, Codex on GPT-6.1 Sol, Deep Research, Canvas, Tasks, Custom GPT creation, and Advanced Voice, ad-free, in one subscription.
• Knowledge workers using AI daily for drafting, summarization, analysis, and structured work.
• Users who want to create Custom GPTs for their workflow (creation requires Plus).
• Anyone comparing $20/month tiers across Claude Pro, Gemini AI Pro, and ChatGPT Plus — ChatGPT Plus wins on ecosystem breadth, Custom GPTs, and the Work agent. It does not include Dots, which start at $100.

Pro $100 Tier (Where Dots Start) — Perfect For:

• Anyone who wants a Dot. This is the cheapest tier that includes one, and it is the real reason to cross from $20 to $100.
• Users who want GPT-6 Pro chat access (currently 200 messages/week) without the $200 or $500 commitment.
• Heavy Codex and ChatGPT Work users who need 5x the Plus allowance — which, from October 30, is only half a step below what the $200 tier delivers.
• Anyone currently on $200 Pro who checks their consumption and finds it under 10x Plus. Downgrading here saves $1,200 a year.
• Not users in the EU, Switzerland, or the UK who are buying primarily for Dots — they are not available in those regions yet.

Pro 200 ($200) And Pro 500 ($500) — Perfect For:

• Pro 200: users consuming more than 10x the Plus allowance in Codex and ChatGPT Work — that is the whole test after October 30, when this tier drops from 20x to 10x and from 200 to 100 weekly GPT-6 Pro messages.
• Pro 200, honestly: if you cannot name your current consumption multiple, you are probably better off at $100 until you can.
• Pro 500: users who are genuinely bottlenecked on generation speed. Ultrafast delivers roughly 300 tokens per second in Codex and Work, about 8x, and that is the tier’s actual product.
• Pro 500: teams running long agent sessions where wall-clock time has a real cost — $6,000 a year needs to be answering a measured bottleneck, not a preference for snappiness.

Skip ChatGPT (Or Pair It With) If:

• Video generation is any part of your workflow — ChatGPT no longer does it at all. Sora’s app closed in April and its API on September 24, with no successor. Use Gemini Omni, Google Flow, or Runway.
• Cited research accuracy is your primary need — Perplexity Pro at $20/month has led every published accuracy index in 2026.
• Coding is your dominant workload — Claude at $20 now leads terminal coding outright (Opus 5.5 at 66.4% vs Astra’s 57.9%), and Claude Code beats Codex for production work.
• You’re deep in Google Workspace — Gemini AI Pro at $19.99 offers tighter Docs/Sheets/Gmail integration.
• Real-time X data is core to your work — Grok SuperGrok at $30 is the only chatbot with native X firehose access.
• You need cost-sensitive API workloads — DeepSeek at $0.14-$0.55 per 1M tokens is 5-18x cheaper than the OpenAI API for comparable chat quality.

💬 Community Sentiment (September 2026)

The ChatGPT review sentiment picture across Reddit, Hacker News, and X in September 2026 is the most divided it has been since the GPT-5 launch — not because the models disappointed, but because OpenAI spent the year removing things. GPT-6 Astra and ChatGPT Work landed well. The three shutdowns and the DevDay pricing changes did not.

What Users Consistently Praise

• ChatGPT Work is the most-praised addition of the year. The framing users keep landing on is that it returns artifacts instead of advice — a finished spreadsheet beats a description of one. Usage backs it: 35 million weekly users across Codex and Work, up from about 5 million Codex users in June.
• GPT-6.1 Sol got an unusually warm reception from developers, for a simple reason: near-Astra coding at a fifth of the price makes the cost of a build session drop immediately rather than eventually.
• Dots generated real excitement in the first days after DevDay, particularly the always-on framing and reachability from Slack, Teams, and voice. Enthusiasm is tempered by how new and how gated they are.
• Folding browser capability into ChatGPT itself — tabs, downloads, navigation, account logins — was received better than the Atlas shutdown that forced it.
• Custom GPT marketplace is still cited as the strongest single ChatGPT advantage over competitors on discovery and sharing.
• Codex keeps its reputation as a real upgrade over chat-style coding, even among users who concede Claude Code leads on production work.

Common Complaints

• The three shutdowns are the dominant complaint, and the anger is about process as much as product. Atlas lasted under a year. Sora went from fastest-growing consumer AI app to discontinued in roughly six months. The ChatGPT agent was removed in early August without advance deprecation notice, which is the detail developers keep returning to.
• The October 30 reduction to the $200 tier reads to many subscribers as a mid-contract downgrade. A one-time transition credit has not settled it — the objection is that an advertised allowance was revised downward after purchase.
• The $500 tier drew immediate sticker-shock commentary: 25x the price of Plus, and more than double the previous ceiling, for what is primarily a speed upgrade.
• Dots being unavailable to Pro users in the EU, Switzerland, and the UK is a live grievance among paying subscribers in those regions, who are billed the same as everyone else.
• Nine tiers is materially more confusing than seven. Users still pick a plan and then discover what they needed sat one rung up — and now three of the rungs differ mainly by allowance multiples rather than features.
• Customer support quality still lags the product polish, particularly on billing, account recovery, and Enterprise contract edge cases.

🎥 Watch: ChatGPT GPT-5.2 Code Red Response (Historical Context)

From our archive — the GPT-5.2 launch coverage when OpenAI shipped the “Code Red Response to Gemini 3.” Worth revisiting as context for this quarter: that was OpenAI racing to add capability, where 2026 has been OpenAI deciding what to stop paying for. Click to play.

✅ Final Verdict

The honest ChatGPT review verdict for September 2026: still the right default for the broadest range of users, but it is a narrower recommendation than it was in May and it comes with homework. ChatGPT Plus at $20/month is genuinely strong — GPT-6 Astra through ChatGPT Work, Codex on GPT-6.1 Sol, Deep Research, Custom GPTs, Advanced Voice, ad-free, and 1.2 billion weekly users’ worth of polish. ChatGPT Work is the best thing OpenAI shipped this year, and Dots may be the best thing next year. But the “one subscription covers everything” pitch took real damage: Atlas, Sora, and Operator are all gone, and the video category is now simply absent.

The default recommendation: ChatGPT Plus at $20/month for most consumer subscribers. Step up to $100 Pro if you want a Dot or GPT-6 Pro chat — that is the boundary that means something now. Treat $200 as a tier to justify rather than assume, and check your consumption before October 30, because it is about to deliver half of what it does today; if you are under 10x Plus, downgrade and keep $1,200 a year. Pro 500 at $500 is a speed purchase for people who can name their bottleneck. And budget for a second subscription: Claude for coding, Perplexity for cited research, Gemini Omni for the video ChatGPT no longer makes.

Overall score: 4/5 ⭐⭐⭐⭐ — down from 4.5 in May. The bump factors are real and substantial: GPT-6 Astra, ChatGPT Work at 35 million weekly users, GPT-6.1 Sol cutting coding costs fivefold, and Dots as the only always-on consumer agent on the market. They are outweighed by a year of subtraction. Three products subscribers built workflows on were discontinued, one of them without advance notice. The $200 tier is being reduced after purchase. The video category was abandoned outright. Half a star is the right cost for a platform that is still the best default but has become materially less safe to build a workflow on.

❓ Frequently Asked Questions

Is ChatGPT Plus still worth $20/month in 2026?

Yes — Plus at $20/month is still the dominant default. You get GPT-6 Astra through ChatGPT Work, Codex running GPT-6.1 Sol, Deep Research, Canvas, Tasks, Custom GPT creation, Advanced Voice, and an ad-free experience (Free and Go are both ad-supported). The same $20 buys Claude Pro or Gemini AI Pro, and ChatGPT Plus still wins on ecosystem breadth. Two things changed since May that you should factor in: Plus no longer includes any video generation, because Sora was discontinued, and Plus does not include Dots — those start at the $100 Pro tier.

What’s the difference between ChatGPT Go and Plus?

Go ($8/month) sits between Free and Plus. It runs GPT-5.6 Luna with higher limits than Free, but it is still ad-supported — Plus is the first ad-free tier, which is a change worth knowing if you signed up to Go specifically to escape ads. Go also excludes the Plus feature surface: no Custom GPT creation, no Codex, no ChatGPT Work, and no GPT-6 Astra, which is restricted to paid tiers from Plus upward. Go is right if you simply want more usage than Free; Plus is right if you want the flagship model or any of the bundled tools.

Pro $100 vs $200 vs $500 — which one do I need?

All three include Dots and GPT-6 Pro chat; they differ on allowance and speed. Pro $100 gives 5x the Plus allowance in Codex and ChatGPT Work plus one included Dot — this is the tier that matters, because it is the cheapest with an agent. Pro 200 currently gives 20x, but from October 30 that drops to 10x and its weekly GPT-6 Pro messages fall from 200 to 100, with a one-time transition credit. Pro 500 ($500/month, announced September 29) has the highest allowance plus Ultrafast inference — roughly 300 tokens per second in Codex and Work, about 8x faster. Practical guidance: if your consumption is under 10x Plus, take the $100 tier and save $1,200 a year over $200. Only buy Pro 500 if you can point to generation speed as a measured bottleneck.

Should I switch from ChatGPT to Claude?

Depends on workload, and the coding answer got clearer. Claude at $20/month now leads terminal coding outright — Opus 5.5 posts 66.4% on Terminal-Bench 4.0 against GPT-6 Astra’s 57.9% — and it is cheaper at the frontier on API ($4/$20 versus $10/$50). Claude also wins careful long-form reasoning and writing. ChatGPT Plus wins general use, Custom GPTs, image generation (Claude still has none), mobile polish, and agents: nothing in Claude’s lineup is quite like an always-on Dot. Many professionals carry both. See our full Claude AI review for the detailed comparison.

Should I switch from ChatGPT to Gemini?

If you live in Google Workspace, Gemini AI Pro at $19.99/month is structurally better for that ecosystem — native integration ChatGPT can’t match. Gemini is also cheaper on API (Gemini 3.1 Pro at $2/$12, Gemini 3.8 Flash at $0.75/$3.75) and is now the clear pick if you need video, since Sora’s shutdown left ChatGPT with none. ChatGPT Plus wins on Custom GPTs, mobile polish, Codex, and Dots. One thing to watch on the Gemini side: its headline Pro API rates step up on January 1, 2027. See our Gemini review for the comparison.

What happened to the ChatGPT Atlas browser?

It was discontinued. Atlas launched in October 2025 as OpenAI’s AI-native browser and stopped working on August 9, 2026 — less than a year later. OpenAI folded browser-based agentic capability into ChatGPT and Codex instead, so ChatGPT itself now handles multiple tabs, downloads, improved navigation, and account logins. If you used Atlas, the capability largely survived; the dedicated browser did not. For browser-agent work specifically, OpenAI points users to ChatGPT Work and a separate cloud browser feature.

What’s the difference between Codex and Codex CLI?

Codex is OpenAI’s full-stack software engineering agent, bundled into ChatGPT Plus and Pro. Codex CLI is the standalone open-source MIT-licensed terminal version for developers who prefer the command line, and competes directly with Claude Code and Gemini CLI. Two September changes matter: Codex now runs GPT-6.1 Sol, which delivers coding comparable to GPT-6 Astra at one-fifth the cost, and Pro 500 subscribers get Ultrafast (~300 tokens/second, about 8x). Heavy production coding still lands on Claude Code most often — Opus 5.5 leads Terminal-Bench 4.0 — but Codex is the right pick for OpenAI-ecosystem developers. See our ChatGPT Codex review.

Can ChatGPT still generate video? What happened to Sora?

No. ChatGPT has no video generation as of September 2026. The Sora app shut down on April 26, 2026 and the Sora API followed on September 24. OpenAI has named no successor and reportedly has no plan to integrate video into ChatGPT. Reporting attributes the decision to compute economics — roughly $500,000 a day in rendering costs — alongside deepfake controversies and revenue that did not justify the spend. Sora had reached a million downloads faster than ChatGPT itself before being discontinued about six months later. If video is part of your workflow you need another tool: Gemini Omni, Google Flow, or Runway.

Does ChatGPT have ads in 2026?

Yes, and the answer changed in a way that matters. As of September 2026 both Free and the $8 Go tier are ad-supported — Plus at $20/month is the first ad-free tier, along with every Pro, Business, and Enterprise plan above it. If you subscribed to Go specifically to escape ads, that is no longer what it buys. The cheapest ad-free ChatGPT is Plus.

📚 Related Reading

• Claude AI Review — Opus 5.5 now leads terminal coding
• Gemini Review — for Google ecosystem users and the cheapest frontier API
• Gemini Omni Review — free video generation, the category ChatGPT exited
• Perplexity Review — for research with cited sources
• Grok Review — for real-time X data workflows
• DeepSeek vs ChatGPT — cost-vs-ecosystem flagship comparison
• ChatGPT Codex Review — full Codex deep-dive
• ChatGPT Agent Mode Review — Agent Mode + Operator coverage
• Gemini Gems Review — Custom GPT alternative on Gemini
• Best AI Developer Tools — broader dev-stack landscape
• Best AI Video Editing Tools — where to go now that Sora is gone

Stay Updated on Flagship AI Models

The flagship chatbot space shifts every month. Subscribe for honest reviews of GPT-6 Astra, Claude Opus 5.5, Gemini 3.8 and whatever ships next — plus tier pricing changes, shutdowns, and product family updates — delivered every Thursday at 9 AM EST.

• ✅ Honest Reviews: We actually test these models, not rewrite press releases
• ✅ Model Releases: GPT-6 Astra / Claude Opus 5.5 / Gemini 3.8 covered within days of launch
• ✅ Pricing Changes: Tier restructures tracked as they happen
• ✅ Product Family Updates: ChatGPT Work, Dots, Spaces, Codex — every new surface, and every shutdown
• ✅ No Hype: Just the AI news that matters for your work

Free, unsubscribe anytime.

Last Updated: October 17, 2026

Tools Covered: ChatGPT on Free, Go ($8/mo), Plus ($20/mo), Pro ($100/mo) and Pro 200 ($200/mo). Comparison testing against Claude (Opus 5.5), Gemini AI Pro (3.1 Pro), Perplexity Pro, and Grok SuperGrok. Pro 500 ($500/mo) and Dots were announced at DevDay on September 29, 2026 and are reported from OpenAI’s announcements and launch coverage rather than from extended hands-on use. All tier pricing verified as of September 30, 2026 — note the $200 Pro allowance change takes effect October 30, 2026, so re-verify before publish.

Slug: /chatgpt/ — evergreen brand-level canonical, unchanged. The slug has held since launch and reflects the brand-anchor positioning we recommend for all flagship-product reviews on the site.

Next Review Update: On October 30, 2026, when the $200 Pro allowance change takes effect — or sooner if Dots reach the EU/UK, GPT-6.1 Sol lands in general ChatGPT, or another surface is discontinued.

Have a tool you want us to review? Suggest it here | Questions? Contact us