OpenAI announced GPT-5.6 on June 26, 2026 — and the developer community was clearly ready for it. It's the first meaningful flagship release since GPT-5, and it comes with a structural change in how OpenAI names its models: three tiers called Sol, Terra, and Luna, each targeting a different capability-cost operating point.
That naming choice is more interesting than it sounds. The GPT-4 era gave us a confusing sprawl — GPT-4, GPT-4 Turbo, GPT-4o, GPT-4o mini, o1, o3, o3-mini — and users regularly had to dig into the docs to figure out which one to actually use. Sol/Terra/Luna is OpenAI's apparent attempt to make the tradeoff explicit from the name.
The June 26 announcement was a preview. That part is over: since July 9, 2026 all six variants — gpt-5.6-sol, sol-pro, terra, terra-pro, luna, luna-pro — have been publicly purchasable, with real published per-token pricing. This page is still a first look rather than a hands-on review, because we haven't run it on our own workloads yet — but the access and pricing unknowns are now resolved.
What we know
Three tiers under the GPT-5.6 name, each with a -pro variant, for six models you can call today. Closed source. A 1,050,000-token context window and a 128,000-token output cap across the line, with text, image, and file input returning text.
Pricing per 1M tokens (input / output):
- Sol — $5.00 / $30.00
- Terra — $2.50 / $15.00
- Luna — $1.00 / $6.00
The -pro variants are listed at the same price as their base tier.
What we still don't know
Benchmark numbers — OpenAI hasn't published comparative results against GPT-5 or across the tiers. And what the -pro variants actually change, given they're priced identically to their base models. Fine-tuning options are likewise unannounced.
Who this is for
Developers already in the OpenAI ecosystem, and anyone who wants a very large context window with multimodal input. The tier split is the practical decision: Luna for high-volume work where $1/$6 per 1M tokens matters, Terra as the sensible default, Sol when you want the ceiling and can absorb $30 per 1M output tokens. The 5x spread from Luna to Sol is wide enough that benchmarking your own workload across all three is worth the afternoon.
The honest take
GPT-5.6 has OpenAI's full flagship backing, a structural naming update that's long overdue, and — now — public pricing and open access. The Anthropic flagship Claude Mythos landed the same week and remains the obvious head-to-head. What's missing is performance data: no public benchmarks from OpenAI, and no hands-on numbers from us.
This page will get a real verdict — including a score — once we've tested it. For now: it's buyable, the specs are published, and the tier you pick is a budget decision more than a capability guess.