GPT-5.6 Sol vs Terra vs Luna: Full Comparison (2026)

GPT-5.6 is not a single model. It's three models.

On July 9, 2026, OpenAI released the GPT-5.6 model family. Sol (Sun), Terra (Earth), and Luna (Moon) — each with a distinct role and a different price point. This article provides a complete comparison of all three.

What You'll Learn in This Article

Differences between GPT-5.6 Sol, Terra, and Luna
API pricing comparison (input, output, cached)
ChatGPT plan availability by model
Benchmark scores for each model
How to leverage the 1.05M-token context window
Recommended models by use case
Cost comparison with Claude Fable 5 and DeepSeek V4 Flash

GPT-5.6 Family Overview

GPT-5.6 is composed of three durable tiers: Sol (Sun), Terra (Earth), and Luna (Moon). The generation number (5.6) represents the training epoch, and each tier continues to evolve independently.

Model	Role	In a Nutshell
☀️ Sol	Flagship Powerhouse	SOTA in deep reasoning, agents, and cybersecurity. The smartest — and the most expensive.
🌍 Terra	Balanced Standard	GPT-5.5-level performance at half the price. Default model for ChatGPT Work.
🌙 Luna	Fast & Low-Cost	Ideal for high-volume tasks like classification, extraction, and routing. Fastest and cheapest.

The naming draws from a Sun → Earth → Moon layered structure: the ultra-powerful Sol at the top, the value-oriented Terra in the middle, and the lightweight, speedy Luna rounding it out.

API Pricing Comparison

OpenAI's API pricing varies significantly across models.

Model	Input (per 1M tokens)	Cached Input	Output (per 1M tokens)	Cost Multiplier
☀️ Sol	$5.00	$0.50	$30.00	5x
🌍 Terra	$2.50	$0.25	$15.00	2.5x
🌙 Luna	$1.00	$0.10	$6.00	1x (baseline)

Key Points:

Cache hits give you 90% off input costs
Minimum cache TTL is 30 minutes
All models are available via gpt-5.6-sol / gpt-5.6-terra / gpt-5.6-luna model IDs in the API
The gpt-5.6 alias routes to Sol (the most expensive — be careful!)

Spec Comparison

Spec	Sol	Terra	Luna
Context Window	1,050,000 tokens	1,050,000 tokens	1,050,000 tokens
Max Output Tokens	128,000	128,000	128,000
Knowledge Cutoff	February 16, 2026	February 16, 2026	February 16, 2026
Reasoning Effort	low/medium/high/max	low/medium/high/max	low/medium/high/max
Speed	Fastest (<280ms TTFT)	Moderate	Slow (deep processing)
Best For	Agents · deep reasoning	Daily tasks · RAG	Classification · extraction · summarization

All three models share the same context window and max output tokens. The differences lie in reasoning depth, speed, and cost.

Available Models by ChatGPT Plan

Access to GPT-5.6 models varies across ChatGPT plans. Note the difference between standard ChatGPT chat and ChatGPT Work / Codex (agent) modes.

Standard Chat (ChatGPT)

Plan	Monthly Price	Available Models
Free	$0	GPT-5.5 Instant (no GPT-5.6)
Plus	$20	✅ GPT-5.6 Sol (medium+ effort)
Pro	$100–$200	✅ GPT-5.6 Sol + Sol Pro (highest quality)
Business	Enterprise	✅ GPT-5.6 Sol (medium+ effort)
Enterprise	Enterprise	✅ GPT-5.6 Sol + Sol Pro

ChatGPT Work / Codex (Agent)

Plan	Available Models	Ultra Mode
Free / Go	✅ GPT-5.6 Terra + Luna	❌
Plus	✅ Sol / Terra / Luna (selectable)	Codex only
Pro	✅ Sol / Terra / Luna + max + ultra	✅ Work + Codex
Business	✅ Sol / Terra / Luna + max	❌
Enterprise	✅ Sol / Terra / Luna + max + ultra	✅ All supported

Ultra mode is an advanced feature exclusive to Pro/Enterprise. It runs multiple agents in parallel — 4 agents by default — splitting complex tasks into sub-tasks.

Benchmark Comparison

Benchmark	Sol	Comparison
Agents Last Exam	52.7% (+5.8 vs GPT-5.5)	GPT-5.5: 46.9%
TerminalBench 2.1	88.8% (Sol Ultra: 91.9%)	Claude Mythos 5: 84.3%
SWE-Bench Pro	64.6%	Claude Fable 5: low 60% range
OSWorld 2.0	62.6% (+15.1 vs 5.5)	GPT-5.5: 47.5%
Coding Agent Index	80 (SOTA)	Claude Fable 5: 77.2
ExploitBench	On par with Mythos Preview (1/3 the output tokens)	Dominates on processing efficiency

Sol scored 80 on the Coding Agent Index, surpassing Claude Fable 5 (77.2) and claiming SOTA (State of the Art).

Compared to Claude Fable 5, Sol delivers equal or better performance with less than half the output tokens, half the time, and roughly 1/3 the cost.

Claude Fable 5 & DeepSeek V4 Flash Comparison

Model	Input (per 1M tokens)	Output (per 1M tokens)	Context	Coding Performance
GPT-5.6 Sol	$5.00	$30.00	1.05M	SOTA (80 pts)
GPT-5.6 Terra	$2.50	$15.00	1.05M	GPT-5.5-class
GPT-5.6 Luna	$1.00	$6.00	1.05M	Lightweight tasks
Claude Fable 5	$10.00	$50.00	200K	77.2 pts (behind Sol)
DeepSeek V4 Flash	$0.42	$1.25	128K	Lightweight tasks

Sol vs. Fable 5 cost comparison:

Sol is 60% cheaper on input and 40% cheaper on output than Fable 5
Yet Sol scores higher on the Coding Agent Index
In short: "cheaper and more capable"

Recommended Models by Use Case

☀️ Choose Sol If You…

Need a serious AI agent for development work
Perform cybersecurity analysis
Do deep research or paper analysis
Run complex multi-step workflows
Are on Plus or above — Sol is directly available in ChatGPT

🌍 Choose Terra If You…

Automate daily work tasks (ChatGPT Work)
Want to keep costs down without sacrificing quality
Build RAG (Retrieval-Augmented Generation) pipelines
Run content generation pipelines
Are a Free/Go user — Terra is your upper limit

🌙 Choose Luna If You…

Need high-volume text classification or extraction
Handle first-pass processing (drafting, summarization, routing)
Prioritize cost over latency for batch jobs
Just want to process a massive volume as cheaply as possible

FAQ

Q1: Which model should I pick?

If budget allows, Sol is the clear choice. For the best bang for your buck, go with Terra. For bulk processing, Luna is your answer. ChatGPT Free/Go users are capped at Terra + Luna.

Q2: Can I use GPT-5.6 on ChatGPT's Free plan?

ChatGPT Free defaults to GPT-5.5 Instant. However, the ChatGPT Work desktop app gives you Terra + Luna access.

Q3: Any gotchas when using Sol via the API?

The gpt-5.6 alias routes to Sol. To avoid accidentally calling the most expensive model, explicitly specify gpt-5.6-terra or gpt-5.6-luna.

Q4: How big is the 1.05M-token context window?

It's roughly 750,000 words — equivalent to the entire "War and Peace" novel plus extra. Perfect for long-form contract analysis or large-scale codebase reviews.

Q5: What is Sol's ultra mode?

An advanced feature that runs multiple agents in parallel. By default, 4 agents process tasks simultaneously. Pro/Enterprise only.

Q6: How long will GPT-5.6 be available?

It's designed as a durable tier — each model gets individual updates. Even as Sol ages, you keep the same pricing until a new Sol generation replaces it.

Summary

The GPT-5.6 family represents a strategic shift from OpenAI: abandoning the "one model to rule them all" approach in favor of three purpose-built models you choose based on your needs.

Sol — For maximum AI capability. Available from $20/month on Plus.
Terra — The best value. Handles 99% of daily work.
Luna — Bulk processing champion. Starts at $1 per 1M tokens via API.

Cheaper than Claude Fable 5 and more capable — GPT-5.6 Sol sits firmly at the top of the leaderboard in the AI model wars of late 2026.

This article is based on information as of July 10, 2026. Check the official OpenAI site for the latest pricing.