GPT-6 Astra vs GPT-6.1 Sol vs GPT-6 Luna: what each OpenAI model is for, prices, limits and which to use (October 2026)

GPT-6 Astra vs 6.1 Sol is the choice most people weighing OpenAI's models face, with GPT-6 Luna as the cheap third option. As of October 23, 2026, OpenAI sells Astra as its most capable model at $10 input and $50 output per million tokens, GPT-6.1 Sol as "near-Astra" at a fifth of that, and Luna for high-volume work at $0.10/$0.50. In ChatGPT the picture is messier: the Chat tab doesn't offer GPT-6.1 Sol at all.

This comparison is based on OpenAI's API pricing page, model pages, deprecations page and changelog, its ChatGPT release notes, help center and Codex pricing page, and results published by Artificial Analysis, LMArena and Vals AI, all checked October 23, 2026. For how the four GPT-6 models launched between September 3 and October 8, see our GPT-6 launch coverage.

The short version

• Price: Astra $10/$50, GPT-6.1 Sol $2/$10, Luna $0.10/$0.50 per million input/output tokens. GPT-6.1 Sol's output costs a fifth of Astra's; Luna's costs a hundredth.
• Context: all three list 1,050,000 tokens in and 128,000 out. Prompts over 272,000 input tokens cost more on all of them.
• In ChatGPT: Chat runs GPT-6 Sol on paid plans and Luna on Free and Go. GPT-6.1 Sol is only in Work and Codex. Astra powers the Pro reasoning option and is limited on Plus.
• Independent scores: Astra and GPT-6.1 Sol are within one point on Artificial Analysis's index and about one point apart on Vals AI's. Luna scores well below both.
• GPT-6 Sol (gpt-6-sol) is left out of the pricing page's default view and appears only when the tables are expanded to "All models." Its model page lists prices and shows no deprecation.

What is each GPT-6 model for?

OpenAI's own words, from its model pages:

• GPT-6 Astra (gpt-6-astra, launched September 3, 2026): "Our most capable model for the most demanding work." OpenAI's launch post for GPT-6.1 Sol says Astra "should be used for the most difficult scientific research tasks."
• GPT-6.1 Sol (gpt-6.1-sol, September 29): "Near-Astra performance for complex work at a lower cost," aimed at coding, computer use and professional work. OpenAI suggests comparing it with Astra on your own tasks.
• GPT-6 Luna (gpt-6-luna, September 22): "Our most efficient model for focused, high-volume tasks."

OpenAI's guidance points two ways. Its models overview says "If you're not sure where to start, use GPT-6 Astra." But OpenAI pitches GPT-6.1 Sol as near-Astra at a fifth of the price, and suggests comparing the two on your own tasks.

How much do GPT-6 Astra, GPT-6.1 Sol and Luna cost?

Last verified: October 23, 2026 (developers.openai.com pricing and model pages; data in M24-gpt-6-models-compared.csv)

Rules that change the bill, per the model pages:

• Long prompts: above 272,000 input tokens, the whole request is billed at 2x input and cache rates and 1.5x output.
• Batch and Flex: 50% of Standard. Fast mode: 2x. Ultrafast: 6x, on Astra and GPT-6.1 Sol only. Regional data residency adds 10%.
• Cache writes: 1.25x the input rate for 30-minute caching. OpenAI's changelog added 12-hour caching on October 20, at 2.5x the input rate.

Ultrafast is expensive enough to flip the order: GPT-6.1 Sol Ultrafast output ($60) costs more than standard Astra ($50). Our story on GPT-6.1 Sol Ultrafast covers that tier.

What happened to GPT-6 Sol on the pricing page?

As of October 23, 2026, OpenAI's API pricing page shows only Astra, GPT-6.1 Sol and Luna in its default view. gpt-6-sol appears only when the tables are expanded to "All models," at $2 input, $0.20 cached and $10 output in Standard, with Batch, Flex and Fast rates but no Ultrafast price. The model isn't being retired, either:

• its model page still lists $2 input, $0.20 cached and $10 output, and points to GPT-6.1 Sol as "the newer Sol model";
• the page shows no deprecation notice, and OpenAI's deprecations page lists none for it;
• that deprecations page still names gpt-6-sol as the recommended replacement for gpt-5.1 and gpt-5.3-codex, which shut down on April 1, 2027.

So OpenAI is steering new users to GPT-6.1 Sol while still pointing older migrations at GPT-6 Sol. Moving from gpt-6-sol to gpt-6.1-sol costs the same per token and halves cached input.

What reasoning options does each model have?

Last verified: October 23, 2026

This matters for latency-sensitive apps. Astra and GPT-6.1 Sol always reason. OpenAI's notes say tool calling on GPT-6.1 Sol needs the Responses API, and on Luna, Chat Completions supports function calling only with reasoning set to none.

Which GPT-6 model do you get in ChatGPT?

Last verified: October 23, 2026

OpenAI says the five-hour figures "are not fixed message limits," and Pro plans "currently" have no five-hour limit. Ultrafast uses included allowance at 8x the standard rate. ChatGPT's October 7 release notes say Work and Codex models didn't change with GPT-6's arrival in Chat, and no entry through October 20 mentions GPT-6.1 Sol reaching Chat. Plan prices run from $0 (Free) and $8 (Go) to $20 (Plus) and $100, $200 or $500 (Pro); see ChatGPT plans compared and our Claude vs ChatGPT vs Gemini comparison for the full plan picture.

Is GPT-6 better than GPT-5.6 Sol?

On price, yes for most uses. GPT-5.6 Sol, which ChatGPT's Chat tab used until October 7, is still in the API at $4 input and $20 output per million tokens. Its model page calls that promotional pricing, available "at least through November 21, 2026." That's twice GPT-6.1 Sol's $2/$10. GPT-5.6 Sol has the same 1.05M-token context and an older knowledge cutoff, February 16, 2026. As of October 23, 2026, OpenAI's pricing page lists it in a "Cyber models" section and in the flagship tables' expanded "All models" view, but not in their default view. It has no deprecation notice.

What do independent evaluators say?

Evaluators call the models through the API, usually at maximum effort, not through ChatGPT. They measure different things, so the results below are kept separate.

Artificial Analysis Intelligence Index v4.3.2, GPT-6 models

Index score (higher is better), each model at "Max" effort

GPT-6 Astra (Max)53

GPT-6.1 Sol (Max)52

GPT-6 Sol (Max)48

GPT-6 Luna (Max)38

Show the data

Model	Score
GPT-6 Astra (Max)	53
GPT-6.1 Sol (Max)	52
GPT-6 Sol (Max)	48
GPT-6 Luna (Max)	38

• Artificial Analysis (Intelligence Index v4.3.2, pages checked October 23, 2026): Astra (Max) 53, GPT-6.1 Sol (Max) 52, GPT-6 Sol (Max) 48, Luna (Max) 38. The index is a composite of evaluations the firm chooses. Its pages show no data date.
• LMArena Text Arena (snapshot dated October 20, 2026): GPT-6.1 Sol (Max) ranks 28th at 1479 ±8 on 7,453 votes; Astra (Max) 34th at 1476 ±6; GPT-6 Sol (Max) 62nd at 1459 ±6; Luna (Max) 96th at 1441 ±4. Gemini 4 Argon (High) leads at 1524, marked preliminary. Scores come from crowd votes on paired chat answers, and GPT-6.1 Sol's and Astra's intervals overlap. The page doesn't clearly state whether style control is applied in its default view.
• Vals AI (Vals Index v2.2, updated October 22, 2026): Astra 65.02% (fifth overall, $21.28 per test), GPT-6.1 Sol 63.94% (seventh, $3.61), Luna 52.32% (20th, $0.45). The index covers agentic tasks in finance, coding, legal and tax. Version 2.2 began October 19 and isn't comparable with earlier versions.

Table data for all three evaluators is in M24-gpt-6-independent-results.csv.

All three agree on one thing: Luna sits well below the other two. They disagree on whether Astra or GPT-6.1 Sol leads, and the gaps are small. OpenAI's own benchmark claims, such as GPT-6.1 Sol matching Astra on DeepSWE v1.1 at "roughly one-fifth the cost," are vendor-reported. For cross-vendor results, see Gemini 4 Argon vs Claude Opus 5.5 vs GPT-6.

Which GPT-6 model should you use?

If cost matters more than staying with OpenAI, our cheap, fast models comparison puts Luna against Claude Haiku 5.5, Gemini Flash and open models.

What's confirmed and what may change

• Confirmed (OpenAI pages, October 23, 2026): API prices, context and output limits, knowledge cutoffs, reasoning options, gpt-6-sol's place outside the pricing page's default view and its live model page, where each model runs in ChatGPT, and the Codex usage estimates.
• Third-party, with limits: the Artificial Analysis, LMArena and Vals AI results, each tied to its date and configuration.
• Uncertain or conflicting:
  • when, or whether, GPT-6.1 Sol reaches ChatGPT's Chat tab;
  • whether gpt-6-sol will be formally deprecated;
  • OpenAI's "GPT-5.6 and GPT-6 Pro in ChatGPT" help article still lists GPT-5.6 Sol and Luna as the Chat models, which contradicts the October 7 release notes;
  • the ChatGPT pricing page couldn't be read for this story, so per-plan context windows in the ChatGPT app aren't given here (see the plans comparison);
  • GPT-5.6 Sol's promotional price ends no earlier than November 21, 2026.