---
title: "OpenAI Releases GPT-5.6 Sol, Terra, and Luna: One Generation, Three Named Tiers"
url: https://xcube.enlightcorp.com.tw/en/news/openai-gpt-5-6-sol-terra-luna-model-family
category: "Frontier Models"
source_type: official
published: 2026-07-09
updated: 2026-07-09
lang: en
---

# OpenAI Releases GPT-5.6 Sol, Terra, and Luna: One Generation, Three Named Tiers

Published 2026-07-09 · Updated 2026-07-09 · Official release · Source: [OpenAI API Changelog](https://developers.openai.com/api/docs/changelog)

**Key answer:** GPT-5.6 is an OpenAI model family released on July 9, 2026, split into the flagship GPT-5.6 Sol, the balanced GPT-5.6 Terra, and the low-cost, high-volume GPT-5.6 Luna; it matters because OpenAI now formally separates capability tiers from model generations, so enterprises must choose models per workload rather than simply adopting the latest one.

According to the OpenAI API Changelog, the GPT-5.6 family was released on 2026-07-09 with three model IDs: gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna. OpenAI's model docs describe GPT-5.6 Sol as the flagship for complex professional work and GPT-5.6 Luna as a nano-tier model for cost-sensitive, high-volume workloads. Both list a 1,050,000-token context window, 128,000 max output tokens, a 2026-02-16 knowledge cutoff, and support for the Chat Completions, Responses, and Batch endpoints.

Pricing moved quickly after launch. The changelog records that on 2026-07-30 GPT-5.6 Luna became 80% cheaper and GPT-5.6 Terra 20% cheaper, and that on 2026-08-21 GPT-5.6 Sol was repriced to $4 per million input tokens and $20 per million output tokens. From 2026-08-05, Fast mode also supported long-context requests on all three models.

The significant change is not another version number but the naming scheme. Sol / Terra / Luna decouples capability tier from generation. For enterprises, model selection shifts from 'should we upgrade?' to 'which tier fits each workload?' — high-volume classification or summarization and complex reasoning may belong on different tiers with cost differences of several times.

Caveats: performance claims come from OpenAI itself; this brief cites only specifications and prices verifiable in official documentation and does not repeat unverified benchmarks. The GPT-5.6 Sol docs also state that Assistants, Realtime, and Fine-tuning endpoints are not supported, so systems built on those endpoints cannot simply swap the model ID.

For an enterprise gateway, GPT-5.6 translates into four concrete checks. First, treat gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna as three separate routing targets rather than aliases of one model. Second, make cost reports break down by model ID, since the price changes on 2026-07-30 and 2026-08-21 can shift the bill for the same workload noticeably within about six weeks. Third, confirm whether each call path uses Chat Completions, Responses, or Batch, the three endpoints listed in the official docs. Fourth, plan separately for any application still relying on the Assistants, Realtime, or Fine-tuning endpoints, because the GPT-5.6 Sol model ID cannot simply be swapped in there.

What to watch: whether OpenAI keeps using short-cycle price cuts to drive migration, and whether the tier naming persists. GPT-6, released in September 2026, already reuses the Sol / Luna names, suggesting a durable product structure rather than a one-off rebrand.

## X Cube view

For enterprises running an OpenAI-compatible API as an internal model gateway, GPT-5.6's tiering means the routing layer should send requests to different model IDs by task and track vendor price changes in cost monitoring; configurations pinned to a single model name miss these cost adjustments.

Tags: OpenAI, GPT-5.6, API, Model Tiering, Enterprise AI, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, Responses API, Chat Completions API