GPT-5.6 Tiers Explained: Which of Sol, Terra, Luna Fits

GPT-5.6 Arrives in Three Tiers — And Naming Them Was the Smart Part

OpenAI moved its most capable system yet into general availability on July 9, 2026, and the most useful thing about the GPT-5.6 launch is not a single benchmark — it is the shape of the lineup. Instead of one monolithic model, GPT-5.6 ships as three named capability tiers — Sol, Terra, and Luna — that let teams match intelligence to the job and the budget. The number marks the generation; the names mark durable tiers that can each advance on their own cadence. For anyone building on large language models, that is a genuinely clarifying idea.

• Sol (flagship): $5 input / $30 output per 1M tokens — state-of-the-art coding, knowledge work, cybersecurity, and science
• Terra (balanced): $2.50 input / $15 output — matches the prior GPT-5.5 flagship at roughly 2x lower cost
• Luna (fast + affordable): $1 input / $6 output — strong capability at the lowest price point
• Where it runs: ChatGPT, Codex, and the API, alongside a new agent called ChatGPT Work

What Does Each GPT-5.6 Tier Actually Do Best?

Think of the three tiers as a ladder you climb only as far as the task demands. Luna is the everyday workhorse — classification, extraction, summarization, chat — where speed and price matter more than frontier reasoning. Terra is the sensible default for production features: it reportedly matches the previous generation's flagship while costing about half as much, which is the kind of quiet efficiency win that pays for itself at scale. Sol is the one you reach for when the problem is genuinely hard — multi-step coding agents, deep research, security analysis — and you want the best available reasoning per token.

Because the tiers share an API surface, you can route dynamically: draft with Luna, escalate to Sol only when confidence is low. That routing pattern is the real cost lever, and GPT-5.6 is clearly designed to encourage it.

Why the Efficiency Story Matters More Than the Headline

OpenAI's framing emphasizes doing more with fewer tokens, and the pricing backs it up. GPT-5.6 also introduces more predictable prompt caching — including explicit cache breakpoints and a 30-minute minimum cache lifetime — which is a meaningful quality-of-life upgrade for agent developers who replay long system prompts thousands of times a day. Cheaper cached context means longer tool-use loops stay affordable.

The broader takeaway for our readers following AI model coverage: the frontier is no longer just about the smartest single model. It is about giving builders a clean menu. If you have been running coding agents, GPT-5.6 slots neatly next to open-weight options like the recently expanded Kimi K2.7 Code in GitHub Copilot, and pairs well with the kind of real-time interfaces we saw in ChatGPT's full-duplex voice update. The competition is fierce, the prices keep falling, and developers are the clear winners.

Sources: OpenAI — GPT-5.6 — July 9, 2026; Simon Willison — The new GPT-5.6 family — July 9, 2026; Neowin — July 9, 2026.