ChatGPT is OpenAI's artificial intelligence assistant, used every week by more than one billion people across the vendor's products. The GPT-6 lineup arrived in September 2026 with Astra, OpenAI's most powerful model, followed by Sol and Luna. Complete ecosystem with ChatGPT Work, Codex, Sites and the Webflow app in ChatGPT.
ChatGPT is built by OpenAI, an American company based in San Francisco and founded in 2015, notably by Sam Altman and Greg Brockman. Launched on November 30, 2022, ChatGPT brought conversational assistants to the general public. OpenAI now claims more than one billion weekly active users across its products and 2.5 million business customers, according to a post by its CFO published on September 8, 2026.
Who is it for, exactly? ChatGPT is aimed primarily at:
Our Webflow agency uses ChatGPT alongside Claude, which carries most of our production. We run two Plus subscriptions, for two uses: image generation, for our site visuals and video thumbnails, then UX/UI work with GPT-6 Astra in Codex. Writing, analysis and code go through Claude, covered in our Claude page.
OpenAI renewed its lineup in September. GPT-6 Astra, released on September 3, 2026, is the vendor's most powerful model. GPT-6 Sol and GPT-6 Luna followed on September 22, at half the price of their GPT-5.6 equivalents. In the regular chat, GPT-5.6 Sol nonetheless remains the base model on paid plans. Astra only appears there as GPT-6 Pro, on the Pro, Business and Enterprise plans.
| Model | Released | API pricing | Context | In ChatGPT | Best for |
|---|---|---|---|---|---|
| GPT-6 Astra GPT-6 | Sept. 3, 2026 | $10 / $50 | 1.05M | Work and Codex from Plus, Chat from Pro (GPT-6 Pro) | The most demanding reasoning, interface design, long-running agents |
| GPT-6 Sol GPT-6 | Sept. 22, 2026 | $2 / $10 | 1.05M | Work and Codex from Plus | Everyday use at a controlled cost |
| GPT-6 Luna GPT-6 | Sept. 22, 2026 | $0.10 / $0.50 | 1.05M | Work and Codex, desktop app | High volume, simple tasks |
| GPT-5.6 Sol | July 9, 2026 | $4 / $20 | 1.05M | Chat, base model on paid plans | Conversation, writing, analysis |
| GPT-5.6 Luna | July 9, 2026 | $0.20 / $1.20 | — | Chat, base model on Free and Go | Free usage |
Prices in US dollars per million tokens, input / output. GPT-5.6 Sol at promotional pricing through at least November 21, 2026, then $5 / $30. Source: OpenAI documentation, September 2026.
GPT-6 Astra for the most demanding tasks. It leads the community ranking WebDev Arena for building web interfaces. OpenAI also reports 97.6% on FrontierMath Tier 4, a research-level mathematics test. It remains slow and expensive: Artificial Analysis measures around 54 tokens per second. On a Plus subscription, Codex allows between 5 and 45 Astra messages per five-hour window, according to OpenAI's estimates.
GPT-6 Sol for everyday work in Work and Codex. At $2/$10, it goes head to head with Claude Sonnet 5, at the same price. The Plus quota rises to 15 to 150 messages per five-hour window. GPT-6 Luna, at $0.10/$0.50, takes high-volume, low-stakes work. Details in the official GPT-6 Sol and Luna announcement.
GPT-5.6 Sol in the chat. The picker offers several thinking levels: Instant, Medium, High, Extra High and Pro. Plus stops at High, Pro unlocks every level. Since September 14, 2026, Instant no longer switches to thinking on its own: the choice is up to the user. Source: official ChatGPT documentation.
Ultra mode splits a task across several sub-agents working in parallel, four by default. Max mode gives a single agent more reasoning time. With GPT-5.6, Ultra is available in Work on the Pro and Enterprise plans. In Codex, it is available from Plus. GPT-6 Luna tops out at Max. These modes burn through quotas faster than a single agent.
On July 9, 2026, OpenAI reorganized ChatGPT around three spaces, brought together in a new desktop app for macOS and Windows. Chat keeps the conversation. Work and Codex handle long-running tasks. The app also includes a browser and computer control, with limits depending on the region.
ChatGPT Work is the long-task agent for non-developers. It runs research, works with connected apps and files, then produces documents, spreadsheets, presentations or sites. Since August 25, 2026, it can be triggered by an event, such as a Gmail email, a Slack message or a GitHub pull request. It can also sign in to password-protected sites on the Plus and Pro plans. Work shares its quotas with Codex. Details in the OpenAI help center.
Codex is dedicated to software development. It runs in the desktop app, on the command line, as an IDE extension and in the cloud, but not on the web or on mobile. We run GPT-6 Astra there for our UX/UI work, with access to design references through MCP.
Sites creates, hosts and shares websites, web apps or games from a prompt or a local project. It ships with a SQL database, file storage, analytics and version management. A custom domain is possible if you already own it. It has been in public beta since July 9, 2026 on the Plus, Pro, Business, Enterprise and Edu plans. Sites was not available in the European Economic Area, Switzerland or the United Kingdom at launch. OpenAI has not announced any lifting of that restriction. The feature offers no data residency either. Source: Sites documentation.
For a prototype, Sites is fast. For a company website, a CMS like Webflow keeps the edge on search visibility, editing by the marketing team and control over data.
Plugins and the Webflow app: since July 9, 2026, the Plugin Directory has replaced the App Directory. A plugin bundles skills, apps and templates, with support for the MCP protocol. On September 11, OpenAI announced the retirement of custom GPTs, which will migrate to this format. The Webflow app, released on July 1, 2026, lets you audit a site, create CMS content and analyze site structure from ChatGPT, without consuming Webflow AI credits for now. Our guide to the Webflow app in ChatGPT covers setup and use cases.
One change deserves the attention of web teams. Since August 31, 2026, Work and Codex can directly use the tools a site exposes through WebMCP, inside the app's browser. A site will therefore be able to offer its own actions to agents, beyond its content. We cover this topic in our agentic web strategy engagements.
Discontinued products: the Atlas browser shut down on August 9, 2026, replaced by the browser built into the desktop app and an extension for Chrome, Edge, Brave, Opera and Vivaldi. The Sora app shut down on April 26, 2026, and its API on September 24, 2026.
Six frontier models shipped in three weeks: Claude Fable 5.1 on September 1, Gemini 3.8 Flash on the 2nd, GPT-6 Astra on the 3rd, then Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna on the 22nd. None of them wins every benchmark. Each vendor publishes the results that favor it. The independent leaderboards contradict each other too.
| Model | Vendor | Released | API pricing | Main strength |
|---|---|---|---|---|
| GPT-6 Astra | OpenAI | Sept. 3, 2026 | $10 / $50 | No. 1 on WebDev Arena, 97.6% on FrontierMath Tier 4 according to OpenAI |
| GPT-6 Sol | OpenAI | Sept. 22, 2026 | $2 / $10 | Direct competitor to Claude Sonnet 5, at the same price |
| Claude Opus 5.5 | Anthropic | Sept. 22, 2026 | $4 / $20 | No. 1 on the Artificial Analysis Intelligence Index at launch (58) |
| Claude Fable 5.1 | Anthropic | Sept. 1, 2026 | $10 / $50 | No. 1 on Agent Arena, No. 2 on WebDev Arena behind GPT-6 Astra |
| Claude Sonnet 5 | Anthropic | June 30, 2026 | $2 / $10 | 1M tokens of context, launch price now permanent |
| Gemini 3.8 Flash | Sept. 2, 2026 | $0.75 / $3.75 | The cheapest and fastest in the comparison, natively multimodal |
Gemini 3.8 Flash at introductory pricing through December 31, 2026, then $1.50 / $7.50. Artificial Analysis and Arena rankings recorded on September 23, 2026, before Opus 5.5 appeared on WebDev Arena.
The same ranking changes from one month to the next. In its own GPT-6 Astra announcement, OpenAI reproduces the Artificial Analysis Intelligence Index: Astra scores 61.2 there, behind Claude Fable 5.1 at 65.7. The next version of the index puts both models level, at 53, on a recalibrated scale. Neither figure is wrong, they simply do not measure the same thing.
The listed price misleads just as much. GPT-6 Astra and Claude Fable 5.1 both cost $10/$50 per million tokens. Artificial Analysis nonetheless records $3.26 per task for Astra against $7.63 for Fable 5.1, because the two models do not consume the same volume of tokens. The only measurement that counts for a team remains a test on its own use cases.
We gave both models the same brief: the landing page for ARC, a fictional customer intelligence SaaS, with an instruction to be original and no skill, template or design file. GPT-6 Astra ran in Codex at low effort, Claude Fable 5.1 in Claude Code at medium effort. The test had three stages: a first draft, an analysis drawing on Mobbin's 600,000 UX and UI references through MCP, then a redesign based on that report.
On the first draft, Astra takes a clear lead. In 11 minutes 34 seconds, it delivers interactive dashboards and a polished integrations section. Fable 5.1 takes around fifteen minutes for a less finished result. Both keep the generic typography of AI-generated sites and get the third-party tool logos wrong.
The gap narrows after the Mobbin analysis. Both models flag the same flaws: an H1 that fails to state the promise and no button to book a demo. In the final version, Astra produces interactive assets and a more restrained art direction. Fable 5.1 handles its orange and black palette well and draws its own illustrations. Astra keeps one advantage Claude lacks: native image generation.
Our verdict leans toward Astra. It leads WebDev Arena with 1,793 points against 1,753 for Fable 5.1. Design Arena nonetheless ranks it 1st overall but 16th in the Website category, well behind Fable 5.1 in 6th. The prompt and the context weigh more than the choice of model.
ChatGPT covers our visual needs. Image generation produces the assets for our sites and the thumbnails for our videos. GPT-6 Astra handles UX/UI: on interfaces, it delivers more polished proposals on the first draft, as our test showed. Two Plus subscriptions are enough for these uses, since Codex gives access to Astra on that plan.
Claude carries the rest of our production: Opus 5.5 day to day, Fable 5.1 for the heaviest tasks. To drive Webflow, the ChatGPT app and the Claude connector rely on the same MCP server. The right choice depends on the tool your team already has open all day. Routing by task has replaced picking a single model: which model on which task, at what cost, with what level of human supervision.
Two grids coexist: subscriptions billed per user and the API billed per token. OpenAI lists its prices in US dollars. In France, Go costs €8 and Plus €23 a month, according to prices reported by Siècle Digital when Go launched there in December 2025. The Free and Go plans have shown ads in France since late August 2026, while no paid plan from Plus up shows any. Source: openai.com/chatgpt/pricing.
| Plan | Price | Models | Included |
|---|---|---|---|
| Free | $0 | GPT-5.6 Luna, with the Think button | Web search, projects, limited memory and images, limited Work and Codex on desktop. Ads |
| Go | $8 | GPT-5.6 Luna, with the Think button | Higher limits than Free, voice with video, scheduled tasks. Ads |
| Plus | $20 | GPT-5.6 Sol in Chat. GPT-6 Astra, Sol and Luna in Work and Codex | Sites in beta, deep research, developer mode, connectors, no ads |
| Pro | $100 or $200 | Every GPT-5.6 Sol level. GPT-6 Pro (Astra) in Chat: 50 or 200 messages per week | 5x or 20x the usage of Plus. New $200 subscriptions paused since September 10, 2026 |
| Business | $20 / $25 Standard $100 / $125 Premium |
GPT-6 Pro: 15 messages per month on Standard, 50 per week on Premium | 2 to 200 seats, SSO, no training on your data by default, SOC 2 Type 2 |
| Enterprise | Custom quote | Depends on workspace permissions, Astra disabled by default at launch | SCIM, key management, data residency in 10 regions, configurable retention |
Prices per user per month, in US dollars. Go, Plus and Pro are billed monthly only. Business: first price billed annually, second billed monthly. Source: openai.com/chatgpt/pricing, September 2026.
On the API side, billing is per million tokens. Caching and deferred processing often weigh more than the choice of model.
| Model | Input | Output | Cached input | Batch and Flex | Beyond 272K |
|---|---|---|---|---|---|
| GPT-6 Astra | $10 | $50 | $1 | $5 / $25 | $20 / $75 |
| GPT-6 Sol | $2 | $10 | $0.20 | $1 / $5 | $4 / $15 |
| GPT-6 Luna | $0.10 | $0.50 | $0.01 | $0.05 / $0.25 | $0.20 / $0.75 |
| GPT-5.6 Sol | $4 | $20 | $0.40 | — | — |
Prices per million tokens. Beyond 272K input tokens, the whole request switches to long-context pricing. GPT-5.6 Sol at promotional pricing. Regional data residency adds 10%. Source: OpenAI pricing documentation, September 2026.
On the GPT-6 lineup, cached input costs 10% of the standard price. Batch and Flex modes halve the bill for any processing that does not need an immediate answer. Fast mode, formerly Priority, doubles it in exchange for speed. Details in the OpenAI pricing documentation.
For professional use, yes. At $20 a month, Plus unlocks GPT-6 Astra in Codex and Work, Sites, deep research and an ad-free experience. The free plan stays on GPT-5.6 Luna. Pro becomes relevant once Astra quotas get in the way: Codex then goes from 5 to 45 Astra messages per five-hour window to 25 to 225 on the $100 Pro plan, according to OpenAI's estimates.
For a team, Business makes sense as soon as client data enters the conversations. Conversations are not used for training by default, SSO is included and OpenAI signs a GDPR-compliant data processing agreement. Business does not, however, offer data residency in Europe, which is reserved for Enterprise and Edu.
Image generation remains our main use of ChatGPT. We use it for our site visuals and for our YouTube video thumbnails. The built-in image model, updated on September 8, 2026 with Images 2.5, can edit a supplied image, keep an approved composition and deliver a high-definition final version. The thumbnail for our comparison video went through five versions before its final 3840 × 2160 pixel master. Claude does not generate images, so ChatGPT naturally completes our setup.
Astra now plays a central role in our UX/UI work. We give it a design system written up as a design.md file, then targeted references pulled from the Mobbin library through MCP. The model produces a first coded mockup that we then take into Figma. Without that context, AI-generated sites all look alike, starting with the typography. The result is a starting point for our UX/UI design team, never a final deliverable.
The Webflow app gives ChatGPT the same capabilities as the Claude connector: SEO audits, CMS item creation, metadata updates. Our workflows have run on the Claude connector since February 2026, but they carry over directly to ChatGPT since the MCP server is the same. For a marketing team that already works in ChatGPT, the app removes the main barrier to adoption. The prompts and limits are detailed in our guide to the Webflow app in ChatGPT.
Long-form writing, analysis of large client specifications and production code go through Claude. Opus 5.5 handles the day to day, Fable 5.1 the heaviest files. The Webflow connector is already part of our routine on the Claude side. These uses are covered in our Claude page. This kind of split between models sits at the heart of our AI and automation offering.
GPT-5.6 Sol for everyday conversation, with a thinking level matched to the difficulty. GPT-6 Astra for the most demanding tasks, available in Work and Codex from Plus. GPT-6 Sol for everyday agents at a controlled cost, GPT-6 Luna for high volume through the API.
Partly. The Free plan gives access to GPT-5.6 Luna with usage limits, web search and limited images. It has shown ads in France since late August 2026. GPT-6 Astra, Sites and an ad-free experience require a Plus subscription at $20 a month.
Plus costs $20 a month, or €23 in France according to prices reported in December 2025. Pro comes in two tiers, at $100 and $200 a month, for five or twenty times the usage of Plus. OpenAI paused new $200 subscriptions on September 10, 2026, amid demand for GPT-6 Astra, according to Fortune. Go, Plus and Pro have no annual billing option.
Astra is the most powerful model, at $10/$50 per million tokens. Sol targets everyday use at $2/$10. Luna covers high volume at $0.10/$0.50. All three share a 1.05 million token context and 128,000 output tokens. In ChatGPT, Sol and Luna are only available in Work, Codex and the API.
Astra, by a short head. In our test, it delivers a more polished first draft and it leads WebDev Arena. The gap narrows as soon as the model gets a design system and references. Design Arena even ranks Fable 5.1 ahead in the Website category. The context you give the model matters more than the model itself.
Yes, with Sites, in beta since July 2026: ChatGPT creates, hosts and publishes a site from a prompt. The feature was not available in the European Union at launch. To manage an existing site, the Webflow app in ChatGPT lets you edit the CMS and metadata of a Webflow site straight from the conversation.
On consumer plans, yes by default. Free, Go, Plus and Pro are used for training as long as the "Improve the model for everyone" setting stays on. You can turn it off in settings or through OpenAI's Privacy Portal. Temporary chats are kept for up to 30 days. Business, Enterprise, Edu and the API do not train models on your data by default. Source: OpenAI help center.
Yes, on the Free and Go plans only. OpenAI started a test in the United States on February 9, 2026, then extended ads to 31 European countries, including France, in late August 2026. In the European Union, ads are contextual at launch, with personalization coming later with consent. No ads appear on Plus, Pro, Business, Enterprise, Edu, for minors or in temporary chats.
Both, depending on the task. ChatGPT offers the broadest ecosystem, image generation and, with GPT-6 Astra, the model at the top of WebDev Arena. Claude leads the agentic and knowledge-work benchmarks published by Anthropic, with Opus 5.5 for daily work and Fable 5.1 for the most demanding reasoning. We use ChatGPT for images and UX/UI, Claude for everything else, as covered in our Claude page.
.svg.png){kind=logo}
