GPT-5.6 Parameters: Sol / Terra / Luna

Last updated：2026-08-12· 15 min read

🚀 Quick access

ChatGPT Domestic：Open entry↗
Mirror site：Open mirror↗
Official ChatGPT：chatgpt.com ↗

Last updated: 2026-09-24 · Model IDs and fields follow OpenAI Platform live docs. For the next flagship, see GPT-6 Astra hands-on. For the GPT-6 daily cost-speed tier, see GPT-6 Sol & Luna.

Overview

After GPT-5.6 shipped, you may see Sol, Terra, and Luna variants in ChatGPT or the API. Picking the most expensive—or never switching—wastes time or depth. This guide maps tasks to variants, not memorized marketing names. If your workspace already has GPT-6, read Sol & Luna and Astra hands-on, then shadow-test before moving defaults off 5.6.

What are Sol, Terra, and Luna?

Think of them as default temperaments within the GPT-5.6 generation (exact copy in-product is authoritative):

Variant	Typical bias	When to select
Sol	Balanced understanding & expression	Daily Q&A, office writing, most coding help
Terra	Deeper reasoning chains	Architecture, multi-constraint planning, hard analysis
Luna	Lower latency, lighter responses	High-frequency short tasks, bulk rewrites, quick drafts

We intentionally avoid hard-coding context windows, prices, or benchmark scores—they change. Put “verify in console/pricing page” in your runbooks.

Task matching matrix

Task	First pick	Backup	Why
Email / weekly report / outline	Sol	Luna	Quality balance; Luna when rushing
Long-doc summary	Sol	Terra	Complex inputs benefit from depth
Trade-off / strategy memo	Terra	Sol	Needs staged reasoning
Bulk titles / tags / micro-edits	Luna	Sol	Speed & cost friendly
Large refactor design	Terra	Sol	Design before code
Small syntax fixes	Sol	Luna	Light task
Multi-step agents	Terra + human review	Sol	See GPT-5.6 prompt & agent guide

Switching workflow (ChatGPT UI)

Default Sol until you hit “too shallow” or “too slow.”
Shallow / missed constraints → Terra + explicit acceptance criteria in the prompt.
High-frequency micro tasks → Luna + shorter prompts.
Run Sol vs Terra on the same task once; pick a team default.
Facts, prices, compliance → human verification regardless of variant.

Comparison prompt:

Answer the same task twice—once as Sol, once as Terra.
Task: [describe]
Format: Conclusion → Steps → Risks → Verify list
Label assumptions you had to make.

For API developers

On OpenAI Platform, the model string is what matters (use current GPT-5.6 IDs from docs).

Smoke-test with Sol in dev.
Add structured outputs via JSON schema / Responses API (guide).
Route in production: Luna for light tasks, Terra for heavy, Sol as default.
Monitor p95 latency, token spend, human edit rate—not hype.

Routing sketch (replace model strings with console values):
if task == "quick_rewrite": model = "gpt-5.6-luna"
elif task == "deep_plan": model = "gpt-5.6-terra"
else: model = "gpt-5.6-sol"

Compare with GPT-5.5 / GPT-4o first?

Variant choice assumes you already want GPT-5.6. If unsure about upgrading:

Access from China

Official: chatgpt.com (model list depends on plan)
Domestic: domestic entry
Paths: China complete guide

FAQ

Are Sol/Terra/Luna totally separate models?

Product-wise they’re usually configured variants of one generation—official wording wins. Task-based defaults matter more than taxonomy debates.

Why is Terra slower?

Deeper reasoning typically costs latency. Use Luna or shrink scope when speed dominates.

Wrong API model name?

Requests fail fast. Maintain smoke tests against the live model list.

Can the app auto-pick variants?

Build routing rules in your stack; in ChatGPT UI you switch manually unless a product auto mode exists.

Official resources

Next reading

Action path

Take one real task this week (plan, code, or copy). Run Sol, then switch per the matrix. Record “task → default variant” in your team wiki.