OpenAI just made GPT-5.6 generally available (July 9, 2026), and it's not one model — it's a three-tier family: Sol, Terra and Luna. That's the real story: OpenAI is admitting what everyone building with AI already knows — the era of "one best model" is over, and best fit beats best benchmark. Here's what each tier is, what it costs, and which one you should actually use.
Independent verdicts on brand-new models take a few weeks to settle. Benchmark figures below are OpenAI's own; we'll update as third-party testing lands.
The three tiers at a glance
| Tier | Built for | API price (per 1M tokens) |
|---|---|---|
| Sol | Flagship — hardest reasoning, coding, science | $5 in / $30 out |
| Terra | Balanced everyday work | $2.50 in / $15 out |
| Luna | Fast, cheap, high-volume tasks | $1 in / $6 out |
- Sol is the frontier model — OpenAI reports it scores a record 53.6 on Agents' Last Exam (long-running professional workflows), and it's launching on Cerebras at up to 750 tokens/second, which is blisteringly fast for a flagship.
- Terra is the sweet spot: OpenAI says it matches GPT-5.5-level quality at half the cost — the tier most teams will default to.
- Luna is the volume workhorse: strong enough for classification, extraction, drafting and chat, at a fraction of Sol's price.
Which one should you actually use?
If you use ChatGPT (not the API): you don't pick a tier by name — GPT-5.6 powers your chats and the app routes to the right tier for the task. The upshot: better answers, faster, especially on complex reasoning. Just keep using ChatGPT; you're already on it.
If you're building with the API:
- Prototyping or high-volume, simple tasks? Start on Luna — it's cheap enough to run at scale.
- Most production work? Terra — GPT-5.5 quality at half the price is the obvious default.
- Hardest reasoning, agentic coding, research? Sol — pay up only where the difficulty justifies it.
The pro move is tiering your own app: route easy calls to Luna, escalate to Terra, and reserve Sol for the genuinely hard requests. GPT-5.6 also adds more predictable prompt caching (30-minute cache life, explicit breakpoints), which meaningfully cuts cost on repeated context.
How it compares to Claude and Gemini
OpenAI's benchmarks put Sol ahead of the field on their chosen tests — but "best fit wins" cuts both ways. Claude still wins fans for writing quality, long-context work and agentic coding via Claude Code; Gemini wins on Google-ecosystem integration and multimodal. The honest 2026 reality: the top models are close enough that price, speed and where it fits your workflow matter as much as raw scores. We keep the head-to-head current in ChatGPT vs Claude vs Gemini.
The bottom line
GPT-5.6 isn't a single leap — it's OpenAI giving you a dial: Luna for cheap-and-fast, Terra for everyday, Sol for the hard stuff. For ChatGPT users it's a free upgrade you already have; for builders it's the clearest cost-vs-capability menu OpenAI has shipped. Match the tier to the task and you'll spend less for better results.
New to picking between assistants? Start with ChatGPT vs Claude vs Gemini, or see where each fits in the ChatGPT & Chatbots category.