AI Model Watch
GPT-5.6 Sol, Terra and Luna: the three tiers, pricing, and which one your business should use
OpenAI split GPT-5.6 into three durable capability tiers and gated the first release behind a US government review. Here is what Sol, Terra and Luna actually do, what they cost, and how to pick the right one for real work.
OpenAI changed how it names models with GPT-5.6. The number (5.6) marks the generation. The names, Sol, Terra and Luna, mark durable capability tiers that can each improve on their own cadence. So instead of guessing whether a smaller variant is good enough, you pick a tier by the job in front of you, and that tier keeps its role across future updates.
There is a second, unusual wrinkle: the first release is narrow. OpenAI made the three models available to roughly 20 organizations after sharing the models and its release plans with the US government, part of a new review process for frontier systems. Individual and consumer accounts are not eligible during the preview. A general release is expected in the coming weeks, through the OpenAI API and Codex.
The three GPT-5.6 tiers at a glance
| Tier | Built for | Input / 1M tokens | Output / 1M tokens |
|---|---|---|---|
| Sol (flagship) | Hardest problems: complex coding, vulnerability research, cybersecurity | ~$5.00 | ~$30.00 |
| Terra (balanced) | High-volume business work: support, internal tools, document analysis | ~$2.50 | ~$15.00 |
| Luna (fast) | Routine automation: summaries, drafting, classification, extraction | ~$1.00 | ~$6.00 |
The headline for cost-conscious teams is Terra. OpenAI positions it as competitive with GPT-5.5 while costing about half as much per token. If GPT-5.5 was already carrying your production workloads, Terra is the natural swap once general access opens.
What actually changed versus GPT-5.5
Three things matter for operators, not benchmarks in isolation:
- Tiering is now explicit. You choose Sol, Terra or Luna by task difficulty and volume, rather than reaching for one default model and overpaying on easy work.
- Sol pushes the security frontier. OpenAI describes Sol as its most capable model yet for long-horizon security tasks, including vulnerability research. That capability is exactly why the release is government-gated.
- Terra resets the price of good-enough. Near GPT-5.5 quality at roughly half the cost changes the math on high-volume agent workloads where token spend, not raw intelligence, is the constraint.
Which tier should your business use
Match the tier to the workload, not to the hype. A quick checklist:
- Use Luna when the task is high-volume and low-stakes: email summarization, first-draft copy, tagging, data extraction, routing. Most automation steps live here, and Luna keeps the per-run cost trivial.
- Use Terra when the task needs judgment and touches customers or records: support replies, document analysis, internal tools, multi-step agent workflows. This is the default for production business automation.
- Use Sol only when the problem is genuinely hard: complex refactors, security review, deep research. Reserve it, because at ~$30 output per million tokens it is expensive to run on work Terra or Luna could handle.
The practical pattern is to route: run the cheap tier by default and escalate to a stronger tier only when a step fails a confidence or quality check. That routing discipline, not picking one model for everything, is where the savings live. It is the same principle behind our AI coding agents comparison.
What this means for AI automation
For teams running agents in production, GPT-5.6 does not change the architecture, it changes the cost curve. A tiered lineup lets you build cheaper by default and only pay for frontier reasoning where it earns its keep. That is exactly how we design enterprise AI automation: a cheap model on the common path, a stronger tier on the hard path, with governance and guardrails around both. For the broader business case on this generation, see our note on GPT-5.6 for business.
One caution: do not rebuild your stack around a model you cannot access yet. Until general availability lands, GPT-5.6 is limited to about 20 organizations. Plan the routing now, validate on GPT-5.5 or Terra when it opens, and keep your prompts and evals model-agnostic so the swap is a config change, not a rewrite.
If you want your own content and tools to surface in AI answers as these models roll into search, our guide to Google preferred sources in 2026 covers the citation mechanics.
Frequently asked questions
What is the difference between GPT-5.6 Sol, Terra and Luna?
They are three capability tiers of the same GPT-5.6 generation. Sol is the flagship for the hardest coding and security work. Terra is the balanced tier for high-volume business tasks at roughly half of GPT-5.5's cost. Luna is the fastest and cheapest tier for routine automation like summarizing, drafting and extraction.
How much does GPT-5.6 cost?
Pricing is per million tokens across the three tiers. Sol is around 5 dollars input and 30 dollars output. Terra is around 2.50 input and 15 output. Luna is around 1 dollar input and 6 output. Output tokens cost far more than input, so response length drives most of your bill.
Can my business use GPT-5.6 right now?
Not generally, yet. The initial preview is limited to about 20 organizations that OpenAI selected after sharing its release plans with the US government. Access runs through the OpenAI API and Codex via an account representative. Individual and consumer accounts are excluded. A general release is expected in the coming weeks.
Why is GPT-5.6 gated behind a US government review?
OpenAI shared the models and release plans with the US government as part of a review process for frontier systems. Sol in particular is described as OpenAI's most capable model for security tasks, including vulnerability research, which raises dual-use concerns. The staged, limited rollout is intended to manage that risk before wider access.
Is Terra better than GPT-5.5?
OpenAI positions Terra as competitive with GPT-5.5 on quality while costing roughly half as much per token. For most production workloads that already ran acceptably on GPT-5.5, Terra should deliver similar results at a lower bill once general access opens, which makes it the likely default for business automation.
Which GPT-5.6 tier is best for automation?
Luna handles the bulk of routine automation steps: summaries, drafting, tagging, extraction and routing, at the lowest cost. Terra covers steps that need judgment or touch customers and records. A good design runs Luna by default and escalates to Terra or Sol only when a step needs stronger reasoning.
What is GPT-5.6 Sol used for?
Sol is the flagship tier aimed at the hardest problems: complex coding, deep research, and cybersecurity tasks such as vulnerability research and exploitation. It is the most expensive tier to run, so it is best reserved for genuinely difficult work rather than everyday tasks that Terra or Luna can handle.
Does GPT-5.6 replace GPT-5.5?
Not immediately. During the limited preview, GPT-5.5 remains the practical option for most teams. GPT-5.6 tiers become a real choice at general availability. Because Terra targets GPT-5.5 quality at a lower price, many teams will migrate their GPT-5.5 workloads to Terra once they can access it.
What does the new Sol, Terra, Luna naming mean?
The number, 5.6, identifies the model generation. The names identify durable capability tiers that can each improve on their own schedule. So Sol always means the flagship tier, Terra the balanced tier and Luna the fast tier, even as future updates raise the quality inside each tier.
How should I prepare for GPT-5.6 before I have access?
Keep your prompts, evaluations and tooling model-agnostic so switching models is a configuration change, not a rewrite. Design a routing layer that uses a cheap tier by default and escalates on hard steps. Validate on GPT-5.5 or Terra when it opens, and measure cost per task rather than chasing the top tier.
Is GPT-5.6 available through the API?
During the preview, Sol, Terra and Luna are available through the OpenAI API and Codex, but only to the limited set of approved organizations working with OpenAI through an account representative. Broader API access for other businesses is expected at general availability, planned for the coming weeks.
How does Digiton help with GPT-5.6 model selection?
We design AI automation that routes work across model tiers: a cheap model on the common path and a stronger tier only where it earns its cost, with governance around both. That keeps quality high and spend predictable, whether you run GPT-5.5 today or migrate to GPT-5.6 Terra and Luna at general availability.
Related
Ready to put AI to work?
Book a discovery audit and we will map the highest-ROI AI agents and automations for your business.
Book a discovery audit →