GPT-6.1 Terra vs Sol vs Luna: the difference, tested in English

👁 127 reads

On 26 October 2026 I ran GPT-6.1 Terra and GPT-6.1 Sol through the same 12 tasks written in British English. Both scored 71 out of 72. Terra did it for $0.050, Sol for $0.103. In the API Terra costs half of Sol per input token and half per output token, and that difference showed up almost exactly in my bill.

• Prices per 1M tokens: Luna $0.10 / $0.50, Terra $1 / $5, Sol $2 / $10, Astra $10 / $50.
• My English test: Terra 71/72, Sol 71/72, both lost the same point on code comments.
• In ChatGPT: Terra only in Work and Codex, not in Chat. The pricing table lists Sol and Astra as “limited” on Plus and “extended” on Pro, and Luna on every plan including Free and Go.
• Independent index: Artificial Analysis puts Sol at 51.8 and Terra at 48.1 (max effort).

Luna, Terra, Sol, Astra: four tiers, one family

OpenAI names the GPT-6 tiers after celestial bodies. Luna is the cheap, fast tier, Terra the middle, Sol the strong all-rounder and Astra the flagship. Version numbers move independently: there is a GPT-6.1 Sol and now a GPT-6.1 Terra, while the newest Luna and Astra are still GPT-6. GPT-6.1 Astra was halted by OpenAI and is not coming for now.

The ChatGPT column comes from OpenAI’s pricing page and help centre, read 26 October 2026. The pricing table did not list Terra yet; the help centre says Terra “is not available in Chat”. For the older generation see my GPT-6 Sol and Luna pricing and test.

My test: Terra vs Sol in British English

I tested both models on 26 October through OpenRouter on OpenAI’s Standard endpoint, reasoning effort medium, one attempt per task. The 12 tasks are local, not translated: an appeal against a Nottingham parking PCN, British idioms, a summary of a Cardiff council transport plan, translation traps, VAT maths, UK date formats, an NHS number validator in Python, a logic puzzle, JSON extraction, UK facts, a search through roughly 30,000 tokens of meeting minutes and proofreading. Code, sums, JSON and facts are checked automatically. Writing is scored blind by Gemini 4 Argon for OpenAI models (GPT-6.1 Sol also judges, but never its own family).

The two OpenAI models were indistinguishable on quality. Both wrote a correct, properly formatted PCN challenge, both found the current bike store code, the second-shift name and the right contractor penalty in the long minutes, and both produced an NHS number function that passed every hidden test. Both lost one point for the same reason: a one-line docstring and almost no comments on the modulus 11 steps.

The difference is cost and speed. Terra used 2,704 output tokens against Sol’s 2,884, so the price gap per token went straight to the bill: $0.050 vs $0.103. Terra also finished the 12 tasks in 51 seconds of total response time against 74 for Sol. Requests ran in parallel, so treat times as rough.

English flatters Terra. In the same test in other languages it fell further behind Sol, for example 65 vs 71 in German, where it wrongly declared the logic puzzle unsolvable.

Which one to pick

• Chat in ChatGPT: Terra is not an option there. On Plus the pricing table lists Sol with a “limited” allowance; Free and Go users get Luna.
• Work and Codex on Plus ($20 a month): pick Terra for documents, spreadsheets, forms and routine code changes. OpenAI says the shared allowance stretches further with Terra than with Sol or Astra. Switch to Sol for a tricky refactor; OpenAI itself recommends Sol and Astra for the hardest coding.
• API at volume: 1,000 requests with 2,000 input and 500 output tokens cost $4.50 on Terra, $9.00 on Sol and $0.45 on Luna. Through Batch, Terra drops to $2.25.
• Long documents: Terra and Sol both take 1,050,000 tokens of context, but above 272,000 input tokens Terra’s whole request is billed at 2x input and 1.5x output.

What about Luna?

GPT-6 Luna was not part of my test, so I have no score for it. What I can say: it costs $0.10 per million input tokens and $0.50 per million output, a tenth of Terra, and it is the model Free and Go users get in Chat. For a side-by-side with other small models, see Haiku 5.5 vs GPT-6 Luna vs Gemini 3.8 Flash.

Prices, availability, benchmarks and my full ranking of nine models are in the GPT-6.1 Terra hub.

Sources

• OpenAI: Introducing GPT-6.1 Terra - read 26 October 2026
• OpenAI API pricing - read 26 October 2026
• OpenAI Help: GPT-6 and other models in ChatGPT - read 26 October 2026
• ChatGPT pricing - read 26 October 2026
• Artificial Analysis: GPT-6.1 Terra - read 26 October 2026

// read moreGPT-6.1 Sol: Pricing, Availability & Non-English Writing TestNewsGPT-6.1 Terra API Pricing: Costs, Batch and Test ResultsAI ToolsOpenAI DevDay 2026: Dots, GPT-6.1 Sol, Pro 500 and MoreNews

Kinetyra