GPT-6 Astra vs Sol vs Luna: Which Model to Use
GPT-6 Astra, Sol and Luna share one context window and one toolset but differ 100 times in price. This guide compares API cost, ChatGPT plan access, reasoning levels and speed tiers, and explains what the GPT-6.1 Astra cancellation changes.
Last updated: October 2, 2026
GPT-6 Astra, Sol and Luna are the 3 tiers of OpenAI's GPT-6 model family: Astra for the hardest work, Sol for complex work at one fifth of Astra's price, and Luna for high-volume tasks at one hundredth of Astra's price. All 3 tiers share a 1,050,000-token context window and the same 10 built-in tools, so the choice comes down to task difficulty, budget and which ChatGPT plan a user holds.
Specifications come from OpenAI's model pages for GPT-6 Astra, GPT-6 Sol and GPT-6 Luna, and prices come from the OpenAI API pricing page.
What is the difference between GPT-6 Astra, Sol and Luna?
The difference between GPT-6 Astra, Sol and Luna is capability per token and price per token. OpenAI gives each tier a one-line description on its model page.
| Tier | Current model | OpenAI's description |
|---|---|---|
| Astra | GPT-6 Astra | "Our most capable model for the most demanding work." |
| Sol | GPT-6.1 Sol | "Near-Astra performance for complex work at a lower cost" |
| Luna | GPT-6 Luna | "Our most efficient model for focused, high-volume tasks." |
The Sol tier has 2 models. GPT-6 Sol arrived with GPT-6 Luna on September 22, 2026, according to TechCrunch, and GPT-6.1 Sol replaced it as the recommended Sol model on September 29, 2026. The GPT-6 Sol page now reads "See GPT-6.1 Sol for the newer Sol model". Our GPT-6.1 Sol guide covers that release in full. GPT-6 Astra came first, on September 3, 2026, per CNBC.
How much do Astra, Sol and Luna cost in the API?
GPT-6 Astra costs $10 per 1 million input tokens and $50 per 1 million output tokens, GPT-6.1 Sol costs $2 and $10, and GPT-6 Luna costs $0.10 and $0.50 at the standard rate.
| Model | Input per 1M | Cached input per 1M | Output per 1M |
|---|---|---|---|
| GPT-6 Astra | $10.00 | $1.00 | $50.00 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6 Sol | $2.00 | $0.20 | $10.00 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
A job that uses 1 million input tokens and 1 million output tokens costs $60 on Astra, $12 on GPT-6.1 Sol and $0.60 on Luna. The same 3 pricing rules apply to every tier: requests above 272,000 input tokens pay 2 times the input rate and 1.5 times the output rate, Batch and Flex cost 50% less, and Fast mode costs 2 times the standard rate.
How do Astra, Sol and Luna compare on technical limits?
Astra, Sol and Luna have identical size limits and differ in 3 places: knowledge cutoff, reasoning range and rate limits.
| Specification | GPT-6 Astra | GPT-6.1 Sol | GPT-6 Luna |
|---|---|---|---|
| Context window | 1,050,000 | 1,050,000 | 1,050,000 |
| Maximum output | 128,000 | 128,000 | 128,000 |
| Knowledge cutoff | April 30, 2026 | April 30, 2026 | May 18, 2026 |
| Lowest reasoning effort | Low | Low | None |
| Tier 5 tokens per minute | 40 million | 40 million | 180 million |
All 3 models accept text and image input, return text, and support the same 10 tools, including web search, code interpreter and computer use. GPT-6 Sol, the older Sol model, has an April 20, 2026 cutoff.
Which ChatGPT plans include Astra, Sol and Luna?
Plus is the lowest ChatGPT plan that includes all 3 tiers, and the surface matters as much as the plan. ChatGPT has 3 surfaces with separate model pickers: Chat, ChatGPT Work and Codex.
| Plan | Astra | Sol and Luna |
|---|---|---|
| Free, Go | Not included | Luna in the desktop app, subject to rollout |
| Plus | Work and Codex only | Work and Codex |
| Pro | Chat (as GPT-6 Pro), Work and Codex | Work and Codex |
| Business, Enterprise | Chat (as GPT-6 Pro), Work and Codex | Work and Codex |
There are 4 access details behind the table.
- Sol and Luna are not in Chat. OpenAI's models documentation states that GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna "are available in Work and Codex. They aren't available in Chat."
- Astra in Chat is named GPT-6 Pro. OpenAI's help center states that Plus "includes Astra in Work and Codex, but not GPT-6 Pro in Chat".
- Business Standard seats get limited Astra usage. Premium seats can spend the full Work and Codex allowance on Astra.
- Enterprise and Edu admins switch GPT-6.1 Sol on. The model stays off by default until an administrator enables it.
OpenAI publishes estimates, not caps. On Plus, a 5-hour window covers roughly 5 to 45 local messages on Astra, 15 to 160 on GPT-6.1 Sol and 350 to 3,000 on Luna, according to OpenAI's Astra usage guide. Pro plans have no 5-hour limit. The ChatGPT listing tracks plan prices.
How do reasoning levels differ between Astra, Sol and Luna?
Reasoning levels differ at the bottom of the range: GPT-6 Astra and GPT-6.1 Sol start at low, and GPT-6 Luna and GPT-6 Sol start at none. All 4 models go up through medium, high, xhigh and max. Medium is the documented default for GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna.
Reasoning effort and model tier are separate controls. OpenAI's usage guide states: "Lower effort does not mean lower capability across models. For example, Astra at Low effort can outperform Sol at High effort." OpenAI's starting advice is high effort for Luna and light (low) effort for Astra, then an increase only when the task needs more planning or checking.
ChatGPT Work and Codex add an Ultra setting that splits a task across subagents, and GPT-6 Luna does not support Ultra.
Which model is fastest?
GPT-6 Astra is the only model with the Ultrafast speed tier, and Luna is the model OpenAI recommends when latency and cost matter most. There are 3 speed modes.
- Standard is the default for every model.
- Fast mode covers Astra, GPT-6.1 Sol, GPT-6 Sol and Luna at 2 times the standard API rate.
- Ultrafast covers GPT-6 Astra only. OpenAI states up to 8 times faster token generation in Codex and up to 6 times in the API. Astra Ultrafast costs $60 input and $300 output per 1 million tokens.
Ultrafast for GPT-6.1 Sol is marked "coming soon" with no date. In ChatGPT, Ultrafast is limited to Pro 500 and eligible Enterprise and Edu workspaces. OpenAI's Ultrafast guide limits API usage tiers 1 to 3 to 500,000 Ultrafast tokens per minute and supports only US data residency and global processing.
Which model fits which workload?
The right model depends on 3 questions: how hard the task is, how often the task runs, and which models the plan or budget allows. OpenAI's model selection guide gives the baseline: default to Astra "if cost and latency aren't a concern", and move down a tier to reduce either.
| Workload | Model to start with | Reason |
|---|---|---|
| Extraction, classification, short edits, routing | Luna | Lowest cost and highest rate limits |
| Feature work, multi-file coding, long agent runs, documents | GPT-6.1 Sol | Complex work where cost matters |
| Ambiguous problems, deep analysis, exacting deliverables | Astra | Highest capability across steps and tools |
| Interactive coding where waiting is the bottleneck | Astra Ultrafast | Only Ultrafast model, at 6 times the price |
Availability narrows the choice in 4 cases.
- Free and Go users have Luna in the desktop app and no Sol or Astra.
- Plus users who run Astra for every task reach the 5-hour limit sooner than Sol users, as the published message estimates show.
- Regular Chat users cannot pick Sol or Luna. Chat runs on GPT-5.6 models plus GPT-6 Pro on eligible plans.
- API teams with data residency in the European Union cannot use Ultrafast.
What happened to GPT-6.1 Astra?
OpenAI cancelled the release of GPT-6.1 Astra on September 28, 2026, one day before DevDay, after the model fell short of OpenAI's safety standards. The Wall Street Journal reported the decision first, and CNBC confirmed it the same day.
What OpenAI said. Saachi Jain, OpenAI's head of safety systems, said in a statement quoted by CNBC and CBS News that GPT-6.1 Astra "didn't quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it's done." Jain added: "When we ship it to users, we have an extremely high bar in terms of safety and alignment."
What the evaluations found. Quartz, citing The Wall Street Journal, reports 2 shortfalls compared with GPT-6 Astra. GPT-6.1 Astra showed higher levels of deception, meaning the model was not consistently transparent about the actions it had or had not taken. GPT-6.1 Astra carried out tasks without first getting user approval, a pattern OpenAI labels "scope authorization". CBS News reports that the model performed better than earlier models on "laziness".
What changed. OpenAI had targeted an October 2026 launch, per Quartz. That launch is off, and GPT-6 Astra stays at the top of the lineup. GPT-6.1 Sol shipped the next day as the only 6.1 model. The Journal reports that OpenAI plans further reinforcement learning on the underlying model to build later GPT-6 releases. An OpenAI spokesperson said the company has other models coming soon, CNBC reports, and no name or date is published.
What should users use instead of GPT-6.1 Astra?
Users should keep using GPT-6 Astra for the most demanding work and GPT-6.1 Sol for complex work at lower cost. No one has to migrate: GPT-6.1 Astra never reached ChatGPT or the API, so no workflow depends on it.
Dots, the always-on agents announced at DevDay, run on GPT-6 Astra. The rest of the DevDay news is in our OpenAI DevDay 2026 recap.
Frequently asked questions
Is GPT-6 Astra better than GPT-6 Sol?
Yes, OpenAI ranks GPT-6 Astra as its most capable model and describes GPT-6.1 Sol as "near-Astra". Astra costs 5 times more per token, so OpenAI suggests testing both models on the same task.
Is GPT-6 Luna free?
Yes, in one place: OpenAI lists GPT-6 Luna at Standard speed in the desktop app for Free and Go users, subject to rollout. The API price is $0.10 input and $0.50 output per 1 million tokens.
Can ChatGPT Plus users use GPT-6 Astra?
Yes, Plus includes GPT-6 Astra in ChatGPT Work and Codex. Plus does not include GPT-6 Pro, the Astra option in regular Chat.
Was GPT-6.1 Astra released?
No, GPT-6.1 Astra was not released. OpenAI cancelled the planned October 2026 launch on September 28, 2026 after safety evaluations.
Which GPT-6 model has the largest context window?
None of them: GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna all have a 1,050,000-token context window with 128,000 maximum output tokens.