The complete guide to GPT models (2026-07): GPT-5.6 Sol, Terra, Luna

OpenAI collapsed its flagship lineup into the GPT-5.6 family on 2026-07-09 — three tiers Sol / Terra / Luna that share the same technical envelope. This guide gets you oriented in under five minutes.

1. Snapshot — 2026-07-14

Source: developers.openai.com/api/docs/models (GPT-5.6 family entries), fetched 2026-07-14.

Shared across all three tiers: 1.05M context, 128K max output, knowledge cutoff 2026-02-16, native multimodal input (text/image/audio/file), and reasoning.effort=none / low / medium / high / xhigh / max.

2. Selection decision tree

Walk these in order:

1. Classification / extraction / translation at volume? → gpt-5.6-luna + reasoning={"effort": "none"}
2. Day-to-day chat / writing / long-document QA? → gpt-5.6-terra + reasoning={"effort": "medium"}
3. Long code / multi-step agent / complex planning? → gpt-5.6-sol + reasoning={"effort": "high"} or xhigh
4. Math contest / scientific analysis / max reasoning? → o-series reasoning model (reasoning.effort fixed at max)

To check "is this tier good enough": let the model self-grade output, auto-upgrade unsure requests to Sol, downgrade obviously cheap ones to Luna. The Responses API ships with a model router that does this for you.

3. Example — Responses API + Python

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="gpt-5.6",                    # alias for Sol; swap to gpt-5.6-terra or gpt-5.6-luna
    input=[
        {"role": "system", "content": "You are a senior Python mentor"},
        {"role": "user", "content": "Explain decorators in one sentence."},
    ],
    reasoning={"effort": "medium"},     # none / low / medium / high / xhigh / max
    max_output_tokens=200,
)

print(response.output_text)
print("tokens used:", response.usage.total_tokens)

To switch tier, change the model field:

• Heavy reasoning → model="gpt-5.6-sol" + reasoning={"effort": "xhigh"}
• High-throughput batch → model="gpt-5.6-luna" + reasoning={"effort": "none"}
• Default daily → model="gpt-5.6-terra" + reasoning={"effort": "medium"}

4. Supporting models — keep them out of the main catalog

Deprecation history:

• 2026-05-12: DALL·E 2 / DALL·E 3 and the Realtime API Beta were removed from the API.
• Pre-2026-07 lineage: GPT-5 / GPT-5.5 / GPT-5.4 / GPT-5.3 / GPT-5.2 were each successive iterations; GPT-5.6 is now the public-facing flagship.

5. Migration — coming from older models

If you are still on older models:

1. GPT-4o / GPT-4 Turbo / GPT-4.1 → gpt-5.6-terra (closest behavioural match)
2. o1 / o1 pro / o3-mini / o-series → gpt-5.6-sol + reasoning={"effort": "high"} or xhigh
3. GPT-5 / GPT-5.5 / GPT-5.5 Pro → gpt-5.6-sol
4. DALL·E 3 → GPT Image 2 (DALL·E 3 is now removed)
5. gpt-4o-realtime-preview (old id) → GPT-Realtime-2.1 or 2.1 mini
6. Chat Completions (messages field) → Responses API (input field)

6. Common errors and fixes

• 401: invalid or expired API key — rotate it.
• 429: rate limit hit — exponential backoff retries, or downgrade from Sol to Terra / Luna.
• 400: bad request. Common causes: passing messages to the Responses API (use input), or oversized images in Sol's context. Resize the image, trim input.
• Poor quality output: bump reasoning.effort from low to medium / high, or switch to Sol. Trim the prompt.
• Too slow: drop Sol to Terra; or keep Sol but lower reasoning.effort to low.

7. Next steps

After this article, read:

• The OpenAI Models Release Notes (living document)
• OpenAI API Beginner: Your First GPT-5.6 Call
• ChatGPT Complete Guide (2026): From Beginner to Expert

Data source and fetch date: GPT-5.6 family descriptions, pricing, context, and knowledge cutoff on this page come from developers.openai.com/api/docs/models and developers.openai.com/api/docs/changelog, fetched on 2026-07-14. We re-fetch and update dataSourcedAt whenever there is a material change; when in doubt, the official pages win.