GPT-5.6 Pricing: Sol vs Terra vs Luna API Costs
GPT-5.6 pricing is simple to read once you know it ships as three model sizes, not one. OpenAI charges per 1 million tokens: Sol at $5.00 input and $30.00 output, Terra at $2.50 and $15.00, and Luna at $1.00 and $6.00. The harder question is which tier fits your workload and how the reworked prompt caching changes your real bill. This breakdown walks through every number, verified against OpenAI's own preview documentation and system card.
Read this first: GPT-5.6 launched June 25, 2026 as a limited preview through the OpenAI API and Codex only. It is not in ChatGPT (Free, Plus, or Pro) during the preview, and access is limited to approved organizations working with an OpenAI account representative. Prices below are preview prices and may change at general availability.
GPT-5.6 Pricing at a Glance
All GPT-5.6 pricing is metered per 1 million tokens, split into input (what you send) and output (what the model returns). Output always costs six times the input rate across all three tiers, which mirrors OpenAI's earlier families and matters because chat and agent workloads that generate long responses are dominated by output cost. Here are the headline numbers.
Put governance around how your team uses AI. The AI Acceptable Use Policy: a deploy-ready template that sets the rules for AI use.
Your purchase helps keep our hubs free to read.
The gap between tiers is large. Luna costs one fifth of Sol on input and one fifth on output, so routing the right requests to the right model is the single biggest lever on your GPT-5.6 API cost. The full table below adds the two caching rates, which we explain in detail further down.
| Model | Model ID | Input / 1M | Output / 1M | Cache write (1.25x) | Cache read (90% off) |
|---|---|---|---|---|---|
| GPT-5.6 Sol | gpt-5.6-sol | $5.00 | $30.00 | $6.25 | $0.50 |
| GPT-5.6 Terra | gpt-5.6-terra | $2.50 | $15.00 | $3.13 | $0.25 |
| GPT-5.6 Luna | gpt-5.6-luna | $1.00 | $6.00 | $1.25 | $0.10 |
If you are also weighing the consumer subscription tiers, see our ChatGPT pricing guide, which covers Plus, Pro, and Business plans. This article is about the GPT-5.6 API tiers specifically. If cost is your deciding factor, our roundup of the cheapest frontier AI APIs ranks GPT-5.6 against rival providers per token.
GPT-5.6 Pricing by Tier: Sol vs Terra vs Luna
OpenAI positions the three GPT-5.6 models by capability and cost rather than by feature toggles. They share the same family and the same preview access rules, so the choice comes down to how much intelligence a job needs versus what you are willing to pay per token. For a plain-English overview of the release itself, see our breakdown of what GPT-5.6 is.
GPT-5.6 Sol: the flagship
Sol is the most capable model in the family and OpenAI's stated flagship. At $5.00 input and $30.00 output per 1M tokens it is the most expensive of the three, priced identically to the previous-generation GPT-5.5. You pay flagship rates for the top of OpenAI's capability curve on software engineering, computer use, scientific research, and cybersecurity work.
GPT-5.6 Terra: the balanced option
Terra is the value story of this release. At $2.50 input and $15.00 output it is exactly half the price of Sol and of GPT-5.5, while OpenAI reports competitive performance to GPT-5.5. For teams that ran GPT-5.5 in production, Terra is the natural drop-in: similar quality on OpenAI's benchmarks at twice the token efficiency. That is why most cost-modeling exercises start with Terra as the baseline and only escalate to Sol where the extra capability pays for itself.
GPT-5.6 Luna: the fast, low-cost tier
Luna is the fastest and most cost-efficient model, at $1.00 input and $6.00 output. It is built for high-volume, latency-sensitive work where you would rather run many cheap calls than a few expensive ones: classification, extraction, routing, and first-pass drafting. OpenAI also noted that Sol will run on Cerebras hardware at up to 750 tokens per second in July for select customers, a separate speed story from Luna's price story.
A spreadsheet that models monthly Sol, Terra, and Luna spend across your own request mix, with cache-hit ratios and a break-even calculator for when to escalate from Terra to Sol. Built to pair with the live calculator below.
Open the cost calculatorHow GPT-5.6 Pricing Changes with Prompt Caching
Prompt caching lets you reuse a shared prefix (a system prompt, a long document, a tool schema) across many requests without paying full input price every time. GPT-5.6 changes how that works, and the changes are billable, so they belong in any cost model.
Earlier families cached automatically through prefix matching, with silent evictions when GPU memory filled. Retention could stretch to 24 hours in good conditions but was never guaranteed. GPT-5.6 replaces that with a predictable, two-sided model:
The practical rule: caching pays off when a prefix is read many times inside its 30-minute window. The 1.25x write premium is recovered after roughly two cached reads, because each read saves 90% of input cost. Beyond that, every reuse is nearly free. Estimate your own numbers with the calculator below.
Estimate only. It treats cached input at the 90% cache-read rate and the rest at the uncached input rate, and ignores one-time cache-write cost, which is small for high-reuse prefixes. Verify current pricing before you commit spend.
How GPT-5.6 Pricing Compares to GPT-5.5
The cleanest comparison is against GPT-5.5, because OpenAI documents it directly. GPT-5.5 is priced at $5.00 input and $30.00 output per 1M tokens, the same as GPT-5.6 Sol. So the headline is not that the flagship got cheaper: it did not. The savings come from Terra, which delivers competitive performance to GPT-5.5 at exactly half the price.
What about Claude and Gemini? We hold our editorial line here: their current API prices are not part of OpenAI's own documentation, so we do not state them as fact in a pricing piece. For a like-for-like model comparison, read ChatGPT vs Claude, and check each vendor's own pricing page before you build a cross-provider cost model. For the backstory on why GPT-5.6 shipped as an API preview rather than a broad launch, see our note on the GPT-5.6 government-coordinated release.
Which GPT-5.6 Model Should You Use
There is no single right answer, only a right answer per request type. The cost-efficient pattern is to route: send cheap, high-volume work to Luna, run everyday production on Terra, and reserve Sol for the hardest jobs where its lead actually changes the outcome. These personas map the three tiers to real roles.
Classification, extraction, tagging, and routing at scale, where latency and unit cost dominate. Luna's $1 input and $6 output keep per-request cost low, and caching a shared instruction prefix cuts it further.
Start with LunaAssistant features, agent loops, and coding help that ran on GPT-5.5. Terra matches GPT-5.5-class quality on OpenAI's benchmarks at half the token cost, making it the default migration target.
Default to TerraLong-horizon coding, computer use, and scientific work where the top of the capability curve pays for itself. Sol sets the state of the art on TerminalBench 2.1 and is worth its flagship price for these jobs.
Escalate to SolOne access caveat shapes all of this: during the preview you can only use these models if your organization has been approved for the API, Codex, or both. Approval for one does not include the other, so confirm your scope with your OpenAI account representative before you design around a specific surface.
Test Your GPT-5.6 Pricing Knowledge
Pick a tier and check your understanding of GPT-5.6 costs and caching. Quick is two questions, Deep is four, and Mastery is all six.
Frequently Asked Questions
Go Deeper
Resources from across Tech Jacks Solutions
GPT, GPT-5.6, ChatGPT, Codex, and OpenAI are trademarks of OpenAI. Claude is a trademark of Anthropic. Gemini is a trademark of Google. Cerebras is a trademark of Cerebras Systems. All product names, logos, and brand identifiers are the property of their respective owners. Tech Jacks Solutions has no commercial relationship with OpenAI. This article is editorially independent.