GPT-5.6 Model Tiers: Sol, Terra and Luna Explained

GPT-5.6 Sol, Terra and Luna were introduced as different capability and efficiency tiers. This guide explains the family historically and shows how to check current access without treating a July 2026 price or app interface as permanent.

GPT-5.6 model tiers comparison graphic showing differences between Luna and Sol.

The tiers as introduced

OpenAI’s June preview described Sol as the flagship, Terra as a balanced model and Luna as a faster, lower-cost option. Its July 30 update reduced Luna’s API prices by 80% and Terra’s by 20%. Those were dated price changes; they do not establish the current cheapest model or the model selected in every ChatGPT account.

TierOriginal positioningQuestion for your workload
SolCapability-orientedDoes it resolve difficult cases enough better to justify its cost and latency?
TerraBalancedDoes it meet the required quality at an acceptable complete-workflow cost?
LunaEfficiency-orientedIs the task narrow enough that the cheaper option still performs reliably?

Check the surface, not just the name

ChatGPT, ChatGPT Work, Codex and the API are different product surfaces. Model availability, selection controls and billing can differ. Check the current model catalog and your account rather than assuming that an assistant’s product name identifies its underlying model.

For API work, record the model identifier, processing mode, input/output prices and any caching assumptions. For a subscription, inspect the plan’s allowances and notices. A historical percentage price cut is not the same as a reduction in the subscription fee.

How to choose without a universal ranking

  1. Collect representative tasks, including difficult and ambiguous cases.
  2. Define a scoring rule before comparing outputs.
  3. Measure correctness, latency, retries and human review.
  4. Record the exact model, settings and date.
  5. Choose the least expensive configuration that meets the quality target.

If a cheaper model needs repeated retries or more review, its nominal token price can mislead. Conversely, paying for the strongest tier for a simple extraction task may add cost without improving the result you need.

What this article no longer claims

We have removed unsupported head-to-head benchmark numbers and blanket statements about which tier a user can manually select in every interface. No model is recommended here for a high-stakes task solely because of its tier name. Use current documentation and an evaluation matched to your application.

Sources and scope

Source check: November 17, 2026. Separated historical tier positioning and price changes from current availability; removed unverified benchmark and interface claims. This is source-based reporting and editorial analysis, not a hands-on product or investment test.

Related reading

Related AI Articles

Leave a Reply

Your email address will not be published. Required fields are marked *