Choosing GPT-5.6 Sol, Terra, or Luna: API costs and ChatGPT access
Compare GPT-5.6 Sol, Terra, and Luna with a fixed extraction task and an API cost example. Check ChatGPT and Codex access separately from API pricing.
3 min read
When choosing GPT-5.6 Sol, Terra, or Luna, identify the product first. API model IDs and metered prices are separate from the options shown in ChatGPT and the usage limits in Codex. An API listing does not establish access in your ChatGPT account.
Which model to try first
The OpenAI model catalog positions Sol for demanding work, Terra for a balance of capability, speed, and cost, and Luna for smaller, faster tasks. These are product positions, not benchmarks measured by this site.
| Task conditions | Starting candidate | Failure that warrants reconsideration |
|---|---|---|
| Extract dates and amounts from short text | Luna | Repeated omissions or format violations |
| Routine changes across several files | Terra | Missed dependencies and failing tests |
| Design or diagnosis with interacting constraints | Sol | Supply missing evidence before changing models |
Prepare the material and acceptance criteria first. More reasoning cannot retrieve a missing log or make an unavailable PDF readable.
API prices and a worked calculation
On September 6, 2026, the official catalog listed these standard text prices per million tokens. Check current pricing before running a job; contracts, processing modes, and additional tools can affect cost.
| Model | Input, USD | Output, USD |
|---|---|---|
| GPT-5.6 Sol | 4.00 | 20.00 |
| GPT-5.6 Terra | 2.00 | 12.00 |
| GPT-5.6 Luna | 0.20 | 1.20 |
For an illustrative request with 10,000 ordinary input tokens and 2,000 billable output tokens, Terra costs 10,000 / 1,000,000 × 2 + 2,000 / 1,000,000 × 12 = $0.044. This is not an observed bill. It excludes cache discounts, tool charges, tax, and retries. Check API usage and billing for actual usage, including reasoning tokens. This calculation does not describe a ChatGPT subscription.
Compare with the same task
Use this fictional sample without personal data:
Input: April 1 Books ¥1,200; April 2 Transport ¥800; unknown date Miscellaneous ¥500.
Output CSV columns: date,category,amount. Leave unknown dates blank.
State the total separately. Do not invent a year or other missing information.
Check for three rows, a ¥2,500 total, a blank unknown date, and no invented year. Keep the input, output constraints, and environment constant. Record correctness, elapsed time, retries, and actual cost for API runs. One success does not establish superiority across tasks.
When a model is missing or a limit is reached
In ChatGPT or Codex, inspect the signed-in account, workspace, model selector, and displayed limit and reset information. Check current account eligibility before buying a different subscription based on an old plan table. The Codex setup guide also helps identify the environment.
If a limit interrupts work, save completed changes, outstanding tests, and the next step. For API errors, investigate model access, the request, and billing limits separately. If you move to a smaller model, validate the result against the same acceptance criteria before adopting it.
Primary sources checked
Important claims should also link to the relevant source in the article body.
- OpenAI model catalogdevelopers.openai.com · official-documentation · Checked: 2026-09-05
- Codex setup guidedevelopers.openai.com · official-documentation · Checked: 2026-09-05