AI and security briefsModels

Which GPT-6 model should I use: Terra, Sol, Astra or Luna?

By Brandon Da Costa, founder of Digiton Dynamics

Published 28 October 2026Updated 28 October 2026Editorial policy

Pick Terra for image-heavy work, screen-driven computer use and drafting. Pick Sol for complex coding, Astra for the hardest tasks and Luna for high-volume jobs. OpenAI released Terra on 26 October 2026. The docs we could read give no benchmark scores for it.

Who should act: Teams that call OpenAI models from production code, or through a gateway such as LiteLLM, and must decide which of the GPT-6 names goes in the model field.

What changed

OpenAI released GPT-6.1 Terra on 26 October 2026 (OpenAI API changelog, 26 October). Its model page calls it a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing, with the ID gpt-6.1-terra. It takes text and images and returns text. Audio and video are unsupported (OpenAI model page, 28 October).

The limits on that page are a 1,050,000 token context window, 922,000 maximum input tokens, 128,000 maximum output tokens and a knowledge cutoff of Jun 01, 2026. Reasoning effort accepts none, low, medium (the default), high, xhigh and max. Tool calling needs the Responses API (OpenAI model page, 28 October).

The pricing page lists Standard rates per million tokens, input then output (OpenAI pricing page, 28 October).

• GPT-6 Astra: $10 and $50.
• GPT-6.1 Sol: $2 and $10.
• GPT-6.1 Terra: $1 and $5.
• GPT-6 Luna: $0.1 and $0.5.

Prompts above 272K input tokens cost 2x on input and 1.5x on output for the whole request. Batch and Flex are 50% lower and Fast mode is 2x (OpenAI model pages, 28 October).

OpenAI's model catalog says to start with Astra for complex reasoning and coding, Sol to balance intelligence and cost, and Luna for cost-sensitive, high-volume work (OpenAI models page, 28 October). Its GPT-6 guide describes Sol as near-Astra performance at a lower cost. None of those pages give a benchmark score for Terra. The announcement page returned HTTP 403 to every fetcher we tried, so any numbers on it are unread here.

Who it affects

• Teams on GPT-5.6 Terra. The pricing page lists that model at $2 and $12 per million tokens, so 6.1 is listed at half the input price and $5 for output (OpenAI pricing page, 28 October).
• Anyone who sends tool calls through Chat Completions. OpenAI's guide says tool calling needs Responses on Astra, Sol and Terra (OpenAI GPT-6 guide, 28 October).
• Code that sets temperature, top_p or top_logprobs. The guide says to remove them whenever reasoning effort is not none.
• Anyone who needs audio, Realtime, embeddings or fine-tuning. The Terra page marks all four as not supported.
• Configs that name gpt-6.1-luna or gpt-6.1-astra. Neither ID is in OpenAI's catalog. The names are GPT-6 Luna and GPT-6 Astra.

What to do now

1. Match the model to the job using OpenAI's own wording. Terra is for visual reasoning, computer use and writing. Sol is for complex coding and professional work. Astra is for the most demanding tasks. Luna is for focused, high-volume work.

   gpt-6.1-terra  gpt-6.1-sol  gpt-6-astra  gpt-6-luna

2. Replay 50 to 100 of your own prompts and images through Terra and the model you run today, at the same reasoning effort. OpenAI gives this advice for the move from GPT-5.6 Terra. Score accuracy, latency and cost per task.
3. Start at medium effort and raise it only when quality gains justify the extra time and tokens, as OpenAI's guide suggests.

   client.responses.create(model="gpt-6.1-terra", reasoning={"effort": "medium"}, input=prompt)

4. Find calls that need changing before you switch.

   grep -rnE "chat\.completions|temperature|top_p" src/

5. Do the cost arithmetic on your real traffic. Our sum, not OpenAI's: a call with 100,000 input and 5,000 output tokens is listed at about $0.125 on Terra and $0.25 on Sol.

What to watch

• Benchmarks. OpenAI's docs publish none for Terra. Read the announcement in a browser before you quote a score, and check it against your own replay.
• Whether Terra gets Ultrafast mode. OpenAI's Ultrafast page calls it broadly available for Astra and Sol only (OpenAI Ultrafast guide, 28 October).
• The claim that Sol delivers near-Astra performance. It is OpenAI's wording and it comes with no figure in the pages we read.

Questions people ask

Digiton's production review checks which model each service calls and what a wrong choice would cost, and the checks are on the vibe coded app audit page.

How this brief was made: researched from the sources below, drafted with AI assistance and checked against each source on the dates shown. See the editorial policy.

Sources

1. OpenAI API changelog, 26 October 2026, read 28 October 2026. https://platform.openai.com/docs/changelog
2. OpenAI model page for GPT-6.1 Terra, read 28 October 2026. https://platform.openai.com/docs/models/gpt-6.1-terra
3. OpenAI model page for GPT-6.1 Sol, read 28 October 2026. https://platform.openai.com/docs/models/gpt-6.1-sol
4. OpenAI model page for GPT-6 Astra, read 28 October 2026. https://platform.openai.com/docs/models/gpt-6-astra
5. OpenAI model page for GPT-6 Luna, read 28 October 2026. https://platform.openai.com/docs/models/gpt-6-luna
6. OpenAI model catalog, read 28 October 2026. https://platform.openai.com/docs/models
7. OpenAI pricing page, read 28 October 2026. https://platform.openai.com/docs/pricing
8. OpenAI guide to using GPT-6, with migration notes, read 28 October 2026. https://platform.openai.com/docs/guides/latest-model
9. OpenAI Ultrafast mode guide, read 28 October 2026. https://platform.openai.com/docs/guides/ultrafast-mode

More briefs

How these briefs are sourced and corrected. All AI and security briefs