Since October 7, 2026, ChatGPT’s Chat tab runs on GPT-6 Sol for Plus, Pro, Business and Enterprise users and on GPT-6 Luna for Free and Go users. GPT-6.1 Sol, its upgrade, powers ChatGPT Work and Codex. GPT-6 Astra is still OpenAI’s most capable model, used by the Pro reasoning option, Dots and the Ultrafast tier.
The names overlap, so many people search for “ChatGPT 6.1 Astra” or “Sol vs Astra” without knowing which model they are using. Knowing which model sits behind each plan is now part of using AI well at work, and it is the kind of practical judgment the Founderz AI & Innovation program builds.
What you’ll take from this
- Which GPT-6 model you get in ChatGPT on each plan, in one table.
- What Sol, 6.1 Sol, Astra and Luna are each designed for.
- How GPT-6.1 Sol compares with OpenAI’s flagship on published benchmarks and API pricing.
- What Intelligent UI, Pro 500, Ultrafast and Dots change for professionals using ChatGPT at work.
Which GPT-6 model does ChatGPT use now?
ChatGPT has three surfaces that can run different models: Chat, Work and Codex. OpenAI’s October 7 release changed the default in Chat only: “The models powering Work and Codex are not changing as part of this release.” [2]
| ChatGPT plan | Chat tab (from October 7 to 8, 2026) | ChatGPT Work and Codex |
|---|---|---|
| Free and Go | GPT-6 Luna | Not part of the GPT-6.1 Sol rollout |
| Plus | GPT-6 Sol | GPT-6.1 Sol |
| Pro | Sol; the Pro reasoning option uses GPT-6 Astra | GPT-6.1 Sol |
| Business | Sol | GPT-6.1 Sol |
| Enterprise | Sol, depending on workplace admin settings | GPT-6.1 Sol |
Paid plans got GPT-6 in Chat first, and the rollout reached Free and Go the following day. [1] [2] GPT-6.1 Sol is available to all Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and OpenAI’s launch page states that it “is not yet available in Chat.” [3]
So if you open a normal chat on a Plus plan today, you are talking to GPT-6 Sol. If you hand a multi-step workflow to Work or Codex, GPT-6.1 Sol does the work.
GPT-6 Sol and Luna, 6.1 Sol and Astra: what each model is for
OpenAI positions the GPT-6 series as one family with different balances of capability and cost. Its model guide gives each one a clear job and its own use cases. [5]
GPT-6 Astra: the flagship
OpenAI introduced its flagship on September 3, 2026, with improvements in coding, research and computer use, and first rolled out to a limited set of organizations. [1] [4] The model guide calls it “our most intelligent model for the best results” and recommends it for the hardest reasoning work. [5] Our guide to ChatGPT Astra covers its launch benchmarks, safety classification and access in detail.
GPT-6 Sol: the everyday model for paid plans
Sol and Luna arrived in ChatGPT Work and Codex on September 22, 2026. [1] Since October 7, a version tuned for everyday conversation powers Chat on paid plans. [2]
GPT-6.1 Sol: near-Astra intelligence at lower cost
GPT-6.1 Sol arrived on September 29, 2026, in Work and Codex and in the OpenAI API as gpt-6.1-sol. [1] [3] The model guide describes it as “near-Astra intelligence for a fifth of the price” and recommends it for complex coding, research and computer use. [5]
GPT-6 Luna: fast work at scale
Luna is the lightest model in the family: “fast and efficient everyday work at scale.” It suits focused, repeated tasks with a clear goal, such as extracting invoice fields, classifying requests or producing structured summaries. [5] It is the model Free and Go users now get in Chat. [2]
GPT-6.1 Sol vs Astra: benchmark results, coding, computer use and pricing
GPT-6.1 Sol “nearly matches GPT-6 Astra’s intelligence on agentic coding, computer use, and professional work” at one-fifth of the flagship’s standard input and output token prices. [3] Each evaluation below comes from OpenAI’s launch page:
- Software engineering: on DeepSWE v1.1, which tests complex tasks in real codebases, GPT-6.1 Sol beats GPT-6 Sol’s best score by 6.4 percentage points at a lower reasoning effort and cost. That makes it a strong option for coding agents.
- Computer use: on the offline set of OSWorld 2.0, it comes within 2.1 percentage points of the flagship’s score at max effort, at roughly one-seventh of the cost per task.
- Scientific research: on Terminal-Bench Science at max effort, GPT-6.1 Sol costs $5.47 per task on average, against $23.80 for Astra. The flagship still scores highest among the models tested, at 68.1%, and remains the recommendation for the hardest long-horizon research.
- Factual reliability: at low reasoning effort, the share of responses with a factual error falls from 11.4% with GPT-6 Sol to 7.7%, a reduction of about 32%. The prompts in this evaluation were deliberately difficult.
All figures: GPT-6.1 Sol launch page. [3]
API pricing per 1M tokens
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Astra | $10.00 | $1.00 | $50.00 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
Sources: launch page and model guide. [3] [5] The developer model page lists a 1,050,000 token context window, 128,000 maximum output tokens, reasoning effort from low to max, and higher rates for prompts over 272K input tokens. Tool calling needs the Responses API. [6]
Prices are per million tokens, and a token is a small chunk of text, so long documents and long answers use more tokens. For a company, average cost per completed task matters more than price per token. A cheaper model that needs three retries can cost more than a stronger one that gets it right first time.
GPT-6 Sol vs GPT-6.1 Sol: what changed, including cache pricing
GPT-6.1 Sol is an upgrade to GPT-6 Sol, not a separate line. It improves across complex professional tasks and business workflows, with the coding and factual reliability gains listed above. [3]
Prompt caching also got cheaper. Cached input costs $0.10 per million tokens, 95% below the standard input price and 50% below the previous Sol cached rate. [3] Cache writes cost $2.50 per million tokens. [6] For an AI agent that rereads the same long instructions or documents, the cache can cut the bill.
For ChatGPT users, the practical difference is where each one runs. Sol is the Chat model on paid plans. GPT-6.1 Sol handles longer jobs in Work and Codex.
How the GPT-6 series rolled out
- September 3, 2026: Astra introduced for a limited set of organizations. [1]
- September 9: Voice can use GPT-5.6 or the new flagship for harder questions. [1]
- September 22: GPT-6 Sol and Luna arrive in ChatGPT Work and Codex. [1]
- September 29: GPT-6.1 Sol rolls out in Work and Codex, alongside the Pro 500 plan, Ultrafast and Dots. [1]
- October 7 to 8: GPT-6 with Intelligent UI reaches Chat on every plan. [1]
- October 9: Composer predictions in Codex launch in beta, using the flagship or GPT-6.1 Sol. [1]
Intelligent UI explained
Intelligent UI lets ChatGPT answer with fully interactive interfaces instead of only text. The announcement calls it “a significant shift in how people interact with ChatGPT.” Its examples include a bill splitter, a retirement savings calculator, a central limit theorem explainer and a road trip planner. [2]
GPT-6 in Chat can also “interleave thinking with answering,” so it reasons while it responds. For questions that need web search, GPT-6 Instant starts answering 44% sooner, on average, than GPT-5.6 Instant. [2] One limit: the Pro reasoning option, which still runs on the flagship model, does not support Intelligent UI. [1]
Pro 500, Ultrafast and Dots
On September 29, OpenAI launched Pro 500 at $500 per month, the Pro plan with the highest included usage. Among Pro plans, Ultrafast for the flagship is available only with Pro 500: you select it in the model picker to get a faster service tier. [1]
The same day, OpenAI introduced Dots: “always-on agents in ChatGPT that can take on ongoing work.” Each dot has its own cloud computer, runs on Astra and brings results back for your review. Dots are rolling out to eligible Pro and Business Premium users aged 18 and older in supported markets. [1]
GPT-6.1 Sol also gets an Ultrafast mode, with up to 8x faster token generation compared to its standard speed in Codex. [3] For developers, Ultrafast costs six times the standard rate. [6]
What it means for professionals using ChatGPT and Copilot at work
Most professionals do not need to pick a model every time. They do need to know what their plan gives them, because the same prompt can behave differently in Chat, Work and Codex.
- Check your surface: a quick question in Chat runs on the Chat model for your plan. A multi-step research or document workflow in Work runs on GPT-6.1 Sol.
- Check your admin settings: on Enterprise, GPT-6 availability in Chat depends on workplace admin settings. [2]
- Review agent output: Dots and Work can complete longer jobs, but a person still needs to check the result before it reaches a client.
Microsoft has also announced GPT-6.1 Sol in Microsoft Foundry. If your company works with Microsoft Copilot, the question is the same as in ChatGPT: which model handles which task, and who reviews the result. Our comparison of Copilot vs ChatGPT covers how the two assistants fit into daily work.
How to choose between Sol, Luna and Astra for each workflow
| Task | Model that fits |
|---|---|
| Short emails, quick answers, simple summaries | Sol in Chat, or Luna on Free and Go |
| High-volume extraction or classification at scale | Luna |
| Multi-step research, documents, spreadsheets | GPT-6.1 Sol in Work |
| Complex coding and computer-use tasks | GPT-6.1 Sol in Codex |
| The hardest reasoning and research problems | Astra, through Pro reasoning or the API |
| Ongoing tasks that run in the background | Dots, on eligible plans |
Start with the cheapest model that does the job well. Move up only when a real task shows it needs more reasoning, more context or more autonomy. For coding, test GPT-6.1 Sol before paying for the flagship.
Frequently asked questions about GPT-6 Sol, Luna and Astra
What is GPT-6 Sol?
It is the model behind ChatGPT’s Chat tab on Plus, Pro, Business and Enterprise plans since October 7, 2026. It first appeared in ChatGPT Work and Codex on September 22, 2026. [1]
Is GPT-6 Sol free?
No. Free and Go users get Luna in Chat. Sol is for paid plans. [2]
Is Astra better than GPT-6.1 Sol?
Astra is the more capable model, but GPT-6.1 Sol comes close on most professional work at one-fifth of the token price. The flagship still leads on the hardest research tasks. [3]
Can I use GPT-6.1 Sol in Chat?
Not at launch. It runs in ChatGPT Work, Codex and the API, and its launch page says it is not yet available in Chat. [3]
How much does GPT-6.1 Sol cost in the API?
$2.00 per million input tokens, $0.10 per million cached input tokens (cache reads) and $10.00 per million output tokens. [3]
Learn to choose and apply AI models at work
Model names will keep changing. The skill that lasts is knowing which work to delegate, what context to give and where your own judgment still matters. The Founderz AI & Innovation program, developed in collaboration with Microsoft, covers applied AI, ChatGPT, Copilot, agents and automation for professional work.
Sources
- OpenAI Help Center: ChatGPT release notes. Dates and availability for every GPT-6 model, Pro 500, Ultrafast, Dots, Intelligent UI and Composer predictions.
- GPT-6 and Intelligent UI for everyone (October 7, 2026). Chat models by plan, Intelligent UI, rollout.
- Introducing GPT-6.1 Sol. Benchmarks, availability, API pricing, Ultrafast.
- GPT-6 Astra launch announcement. Launch and positioning.
- A model guide for the GPT-6 family (October 2, 2026). What each model is for, pricing.
- GPT-6.1 Sol API model page. Context window, output limit, reasoning effort, pricing modes.
- OpenAI Help Center: ChatGPT release notes, September 22, 2026 entry.
