GPT Image 2.5 is live — OpenAI's newest image model, targeted edits that leave the rest of the frame alone
All Posts
GPT-6 Sol vs Terra vs Luna: Where Astra and 6.1 Sol Fit
2026/10/07

GPT-6 Sol vs Terra vs Luna: Where Astra and 6.1 Sol Fit

GPT-6 Sol vs Terra vs Luna explained: OpenAI's GPT-6 lineup has Astra, Sol, 6.1 Sol and Luna but no GPT-6 Terra. Prices, efforts, scores and which to use.

GPT-6 Sol, Terra and Luna are OpenAI language models, not the Solana or Terra Luna cryptocurrencies. If you searched "GPT-6 Sol vs Terra vs Luna" and got coin prices, some assistants read the query that way. This article is about the API models.

The short version: OpenAI's GPT-6 family has four models in its API catalogue, namely GPT-6 Astra, GPT-6 Sol, GPT-6.1 Sol and GPT-6 Luna. None of them is called Terra[1]. Terra exists only in the previous generation, as GPT-5.6 Terra[2]. Below is one table that places all five, what OpenAI and an independent benchmark say about how they differ, and which one to pick for which job.

TL;DR

  • There is no GPT-6 Terra. OpenAI's model catalogue lists GPT-6 Astra, GPT-6 Sol, GPT-6.1 Sol and GPT-6 Luna, plus GPT-5.6 Sol, Terra and Luna from the July generation[1].
  • Astra is the top tier. OpenAI calls GPT-6 Astra "our most capable model for the most demanding work", at $10 input and $50 output per million tokens[1][3].
  • Sol is the middle tier and already has a successor. GPT-6 Sol shipped on September 22, 2026; GPT-6.1 Sol followed on September 29 at the same $2 input and $10 output, with cheaper cached input[4].
  • Luna is the efficient tier at $0.10 input and $0.50 output, built "for focused, high-volume tasks"[1][3].
  • Scores follow price. On Artificial Analysis's Intelligence Index at max effort, GPT-6 Astra scores 53, GPT-6.1 Sol 52, GPT-6 Sol 48 and GPT-6 Luna 38[5].

The GPT-6 lineup in one table

ModelAPI model idReleasedOpenAI's descriptionReasoning effortInput / cached input / output (per 1M tokens)
GPT-6 Astragpt-6-astraSep 3, 2026Most capable model, for the most demanding worklow to max$10 / $1 / $50
GPT-6.1 Solgpt-6.1-solSep 29, 2026Near-Astra performance for complex work at a lower costlow to max$2 / $0.10 / $10
GPT-6 Solgpt-6-solSep 22, 2026Built for complex coding and agentic workflowsnone to max$2 / $0.20 / $10
GPT-6 Lunagpt-6-lunaSep 22, 2026Most efficient model, for focused, high-volume tasksnone to max$0.10 / $0.01 / $0.50
GPT-5.6 Terragpt-5.6-terraJul 9, 2026Balances intelligence and cost; roughly the old "mini" tiernone to max$2 / $0.20 / $12

Sources: model catalogue and model pages[1], OpenAI API changelog[4], OpenAI Standard pricing for prompts up to 272K input tokens[3], GPT-5.6 Terra model page[2].

All five share the same limits on paper: a 1,050,000-token context window, up to 922,000 input tokens and 128,000 output tokens, text and image input, text output[1]. Prompts above 272K input tokens are billed at 2x the input rate and 1.5x the output rate for the whole request on every model in the table[1]. What separates them is the tier, the price and, at the edges, the API surface.

Is there a GPT-6 Terra?

No, not in OpenAI's API. OpenAI's catalogue has no GPT-6 Terra entry, and the model page you would expect at gpt-6-terra does not exist[1]. The GPT-6 Sol and Luna announcement, the GPT-6.1 Sol announcement and OpenAI's GPT-6 model guide do not mention a Terra either[6][7][8].

The name comes from GPT-5.6. That generation launched on July 9, 2026 as three tiers: GPT-5.6 Sol for frontier capability, GPT-5.6 Terra for a balance of intelligence and cost, and GPT-5.6 Luna for efficient, high-volume work[4]. OpenAI's own pages map them onto the older naming: Sol to the unsuffixed model, Terra to the "mini" tier and Luna to the "nano" tier[2].

GPT-6 kept Sol and Luna and moved the top of the range to a new name, Astra. A popular Reddit thread argues that GPT-6 Sol is effectively Terra renamed, but OpenAI has published nothing that says so. What the price list does show is that GPT-6 Sol now costs the same per input token as GPT-5.6 Terra ($2) and less per output token ($10 against $12)[3]. If you were using Terra for cost reasons, GPT-6 Sol is the natural model to test against it.

GPT-6 Sol and GPT-6.1 Sol are the same tier

GPT-6.1 Sol is an upgrade to GPT-6 Sol, released one week after it[7]. OpenAI's GPT-6 Sol model page now points readers to "GPT-6.1 Sol for the newer Sol model", and the GPT-6 guide lists Astra, 6.1 Sol and Luna as the three current choices[1][8].

The two Sol models share input and output prices. The differences that matter in practice:

  • Cached input: $0.10 per million on GPT-6.1 Sol against $0.20 on GPT-6 Sol[3].
  • Reasoning effort: GPT-6 Sol accepts none; GPT-6.1 Sol does not, and starts at low[8].
  • Tools in Chat Completions: GPT-6 Sol supports function calling there only with effort none; GPT-6.1 Sol needs the Responses API for any tool calling[8].
  • Scores: OpenAI reports GPT-6.1 Sol beating GPT-6 Sol's best DeepSWE score by 6.4 points and matching GPT-6 Astra on that benchmark at roughly a fifth of the cost[7].

If someone tells you "Sol is near-Astra", they are usually quoting the 6.1 announcement. Check which Sol a comparison actually used before you rely on it.

GPT-6 Sol vs Luna: what the scores say

OpenAI says it trained Sol and Luna "with similar methods as GPT-6 Astra" and cut their API prices by 50% against GPT-5.6's promotional pricing[6]. The gap between them is wide on price: Luna's per-token rates are a twentieth of Sol's on input and output[3].

Artificial Analysis ran both at max effort on its own benchmark suite:

Measure (max effort)GPT-6 SolGPT-6 Luna
Intelligence Index4838
Terminal-Bench 4.0 (agentic coding and terminal use)44%13%
AA-LCR (long-context reasoning)84%83%
Cost per task$1.04$0.07
Output speed95 tokens/s128 tokens/s

Source: Artificial Analysis, retrieved October 7, 2026[5].

The pattern is clear. On long-context reading Luna is close to Sol. On multi-step agentic coding in a terminal, it falls well behind. OpenAI's own numbers are friendlier to Luna on coding: on DeepSWE v1.1, Luna at max effort scored 66.6% against Sol's 68.8%[6]. The two benchmarks measure different things, so read them as a range, not a contradiction.

Picking a GPT-6 model by task

OpenAI publishes a ladder of model and effort pairs, ordered from cheapest to most capable. It is the clearest official answer to "which one should I use":

StepModel and effortOpenAI's example work
1Luna, LowFine-grained edits, well-scoped problem-solving, simple data extraction
2Luna, Extra highFinding context across apps, prioritizing work, problems with clear constraints
3GPT-6.1 Sol, MediumComplex technical work and coordinated deliverables you expect to revise
4GPT-6.1 Sol, Extra highPolished deliverables, decisions built from conflicting evidence
5Astra, LowConcise writing and content adaptation that preserve facts and nuance
6Astra, MediumAmbitious projects that need broad context and complete results
7Astra, Extra highDemanding analysis and complex deliverables with exacting requirements

Source: OpenAI model selection guide[9].

OpenAI's guidance is to start low and keep "the lightest setting that meets your quality bar". It also says that if cost and latency are not a concern you can default to Astra[9]. In practice:

  • Classification, extraction, tagging, routing: Luna. OpenAI's own examples are invoice fields, request classification and structured summaries[8].
  • Coding agents, debugging, document-heavy analysis: Sol. OpenAI's guidance now names GPT-6.1 Sol for this tier; GPT-6 Sol remains the option where you need effort none.
  • The hardest reasoning, research and computer-use work: Astra[8].
  • Mixed traffic: route by difficulty, with Luna taking the bulk and Sol or Astra taking what Luna gets wrong.

Calling GPT-6 Sol, Luna and Astra on reAPI

reAPI serves GPT-6 Sol, GPT-6 Luna and GPT-6 Astra through one Chat Completions endpoint and one API key. The model ids are gpt-6-sol, gpt-6-luna and gpt-6-astra. GPT-6.1 Sol is not on reAPI's model list as of October 7, 2026.

A few request details differ between the three on reAPI[10]:

  • Sol and Luna accept all six reasoning_effort values, none through max, with medium as the default. Astra accepts low through max; none returns 400.
  • temperature, top_p, frequency_penalty and presence_penalty return 400 on Sol and Luna. Leave them out.
  • Function calling returned tool_calls at medium effort as well as none on Sol and Luna when measured on September 24, 2026.

Current per-token prices are on each model page. For the cost side in detail, see GPT-6 Sol pricing and GPT-6 Luna API pricing; for the Sol versus Astra decision, see GPT-6 Sol vs Astra.

FAQ

What is GPT-6 Sol and Luna?

They are two OpenAI models released on September 22, 2026. GPT-6 Sol is built for complex coding and agentic workflows; GPT-6 Luna is OpenAI's most efficient model, for focused, high-volume tasks[1][4]. Both take text and images and return text. They sit below GPT-6 Astra in the GPT-6 family.

Where would Terra fit in this new generation?

It doesn't, so far. OpenAI's catalogue has no GPT-6 Terra[1]. Terra is the middle tier of GPT-5.6, and at $2 input GPT-6 Sol now lists at the same input price as GPT-5.6 Terra with a lower output price[3].

Is GPT-6 Luna higher than Terra?

Not on price tier: Luna is the low-cost tier and GPT-5.6 Terra was the middle one. GPT-6 Luna costs $0.10 input and $0.50 output per million tokens against Terra's $2 and $12[3]. OpenAI has not published a direct Luna versus Terra benchmark, so test both on your own tasks if you are moving off Terra.

Is Astra the same as GPT-6?

GPT-6 Astra is one model in the GPT-6 family, the most capable one. The family also includes GPT-6 Sol, GPT-6.1 Sol and GPT-6 Luna[1]. Astra was released first, on September 3, 2026[4].

GPT-6 Luna vs 6.1 Sol

GPT-6.1 Sol costs 20 times more per input and output token than GPT-6 Luna ($2 and $10 against $0.10 and $0.50)[3]. On Artificial Analysis's Intelligence Index at max effort, 6.1 Sol scores 52 and Luna 38[5]. OpenAI's ladder puts Luna at the first two steps and 6.1 Sol at the next two[9].

GPT-6 Sol or Luna for coding?

For agentic coding, Sol. Artificial Analysis measured 44% for Sol and 13% for Luna on Terminal-Bench 4.0 at max effort[5]. Luna is closer on OpenAI's DeepSWE run (66.6% against 68.8%), so it can be worth testing for well-scoped fixes where its lower price matters[6].

GPT-6 Sol vs Terra vs Luna: what to use today

Read the lineup as three tiers, not three names: Astra at the top, Sol in the middle, Luna at the bottom, with GPT-5.6 Terra as the previous generation's middle tier that GPT-6 did not carry over. For high-volume, checkable tasks, start on Luna. For coding agents and hard analysis, use Sol, and compare the 6.1 version where your platform offers it. Keep Astra for work where a wrong answer costs more than the tokens. That is the practical answer to GPT-6 Sol vs Terra vs Luna.

References

  1. OpenAI. Models (API model catalogue) and the GPT-6 Astra, GPT-6 Sol, GPT-6.1 Sol and GPT-6 Luna model pages. Retrieved October 2026 from developers.openai.com/api/docs/models
  2. OpenAI. GPT-5.6 Terra model page. Retrieved October 2026 from developers.openai.com/api/docs/models/gpt-5.6-terra
  3. OpenAI. API pricing. Retrieved October 2026 from developers.openai.com/api/docs/pricing
  4. OpenAI. API changelog. Retrieved October 2026 from developers.openai.com/api/docs/changelog
  5. Artificial Analysis. GPT-6 Sol (Max) vs GPT-6 Luna (Max) and GPT-6.1 Sol (Max) vs GPT-6 Astra (Max): model comparisons. Retrieved October 2026 from artificialanalysis.ai/models/comparisons/gpt-6-sol-vs-gpt-6-luna
  6. OpenAI. Introducing GPT-6 Sol and Luna. Retrieved October 2026 from openai.com/index/introducing-gpt-6-sol-and-luna
  7. OpenAI. Introducing GPT-6.1 Sol. Retrieved October 2026 from openai.com/index/introducing-gpt-6-1-sol
  8. OpenAI. Using GPT-6 (API guide) and A model guide for the GPT-6 family. Retrieved October 2026 from developers.openai.com/api/docs/guides/latest-model and openai.com/index/practical-guide-building-gpt-6
  9. OpenAI. Model selection. Retrieved October 2026 from developers.openai.com/api/docs/guides/model-selection
  10. reAPI. GPT-6 Sol, GPT-6 Luna and GPT-6 Astra API docs. Retrieved October 2026 from reapi.ai/docs/gpt-6-sol, reapi.ai/docs/gpt-6-luna and reapi.ai/docs/gpt-6-astra
All Posts

More Posts