What shipped on July 9
OpenAI didn't ship one new model, it shipped three. Sol, Terra, and Luna all carry the GPT-5.6 label, but the number is just the generation marker. Sol, Terra, and Luna are separate capability tiers, and OpenAI has said each one can get updated on its own schedule going forward.
Pricing, per million tokens:
| Model | Input | Output |
|---|---|---|
| Sol | $5.00 | $30.00 |
| Terra | $2.50 | $15.00 |
| Luna | $1.00 | $6.00 |
Sol is the flagship, built for heavy reasoning, coding, and scientific work. Terra targets close to Sol's quality at less than half the price. Luna is the cheap option: a fifth of Sol's cost on both input and output, meant for high-volume, simpler jobs.
What's new besides the price tags
Two changes matter more than the benchmark charts. Programmatic Tool Calling, new in the Responses API, lets the model write and run small in-memory programs that coordinate tools and process the results themselves. It's Zero Data Retention compatible, which matters if you're working in a regulated industry. Prompt caching also got more predictable: explicit cache breakpoints and a 30-minute minimum cache life mean you can plan around it instead of guessing. A beta multi-agent mode lets GPT-5.6 run several subagents at once and merge their output into one response.
Is it better?
Depends who you ask. If you're running coding agents on GPT-5.5 or an older model, it's worth testing Sol. Early testers say it handles messy, real-world repositories better than the previous generation. But the benchmark numbers OpenAI published are vendor-reported, and a good chunk of the developer community is holding off, waiting for real-world testing before trusting the charts. For how the consumer-facing apps stack up instead of the raw API tiers, see our [ChatGPT Plus vs. Claude vs. Perplexity comparison](https://questloops.com/blog/chatgpt-plus-vs-claude-vs-perplexity).
Which tier should you use
The price gap between Sol and Luna is 5x on both input and output. The quality gap on most everyday tasks is nowhere near 5x. For simple, high-volume work like classification, extraction, or short-form drafting, Luna is the better economic bet. For anything where a wrong answer is expensive, like production code or technical writing that ships to customers, Sol or Terra earn their price. If you're not sure where you land, start with Terra: it's built to land near Sol's quality for under half the cost.
One catch: none of this is in the regular ChatGPT app the way GPT-5.5 was. At launch it's API and Codex access only, gated to a limited partner list. If you're a ChatGPT Plus subscriber hoping to try Sol in the chat window (see our guide on [getting ChatGPT Plus for less](https://questloops.com/blog/how-to-get-chatgpt-plus-free)), there's no confirmed date for that yet.
