Route each request to the least costly GPT-5.6 tier that meets your application’s quality and latency requirements. A sensible starting policy is Luna for simple, well-scoped work, Terra for a balance of capability and cost, and Sol for complex tasks—but treat that mapping as a hypothesis to test, not a guaranteed ranking for your workload.
What separates Sol, Terra, and Luna?
OpenAI positions GPT-5.6 Sol as its flagship for complex professional work, Terra as a balance of intelligence and cost, and Luna for cost-sensitive, high-volume workloads. Those descriptions can guide an initial policy; they do not establish how the models will perform on your particular prompts.
As an Amazon Associate I earn from qualifying purchases.
| Model | OpenAI positioning | Model ID | Standard input / cached input / output per 1 million tokens (USD) |
|---|---|---|---|
| GPT-5.6 Sol | Flagship for complex professional work | gpt-5.6-sol |
$4 / $0.40 / $20 |
| GPT-5.6 Terra | Balances intelligence and cost | gpt-5.6-terra |
$2 / $0.20 / $12 |
| GPT-5.6 Luna | Cost-sensitive, high-volume workloads | gpt-5.6-luna |
$0.20 / $0.02 / $1.20 |
These are the standard text rates currently listed on OpenAI’s Sol, Terra, and Luna model pages, checked October 7, 2026. Prices are in USD and may change; check the live pages before budgeting or deployment. OpenAI announced Terra’s $2 input and $12 output rates and Luna’s $0.20 input and $1.20 output rates as effective July 30, 2026 (OpenAI pricing announcement).
Input and output costs both matter. A request with a long answer can incur substantial output charges, so estimate or measure both token types rather than choosing by input price alone. Cached input has a separate listed rate; use it only when the applicable request actually qualifies for cached-input pricing.
#1 Best Overall
How should you choose a tier for a task?
Start with a workload hypothesis
For a low-cost initial policy, try Luna on routine, constrained tasks; Terra where your work needs a stronger balance; and Sol for complex tasks where evaluation shows its results justify the higher rates. This is an implementation choice inferred from OpenAI’s tier descriptions and pricing, not an official routing recipe.
Set acceptance criteria before comparing
Build a small evaluation set from real requests, including ordinary cases and consequential edge cases. Define what counts as acceptable correctness and what latency the application can tolerate. Send the same inputs to candidate models and compare outcomes alongside token use and response time. OpenAI’s model-selection guidance recommends experimenting on representative tasks and comparing quality with cost; it does not promise that any tier will meet a specific application’s bar: model selection guidance.
Keep the least costly model that clears your defined quality and latency thresholds for each task class. Revisit the decision as your prompts, traffic mix, or requirements change. There is no directly comparable published benchmark in the cited materials for these three models on Python-router workloads, so the price table cannot establish their relative quality or latency.
Build a simple Python router
The Responses API lets a request specify its model through the model parameter. The example below maps a task class supplied by your application to a model ID:
from openai import OpenAI
client = OpenAI()
MODEL_BY_TASK = {
"routine": "gpt-5.6-luna",
"balanced": "gpt-5.6-terra",
"complex": "gpt-5.6-sol",
}
def respond(task_class: str, prompt: str):
model = MODEL_BY_TASK[task_class]
return client.responses.create(model=model, input=prompt)See the Python Responses API reference for the request method and parameters. The mapping is an illustrative policy, not automatic prompt classification: your application must decide the task class. This minimal function also does not assess correctness, handle retries, or guarantee savings. Add validation and error handling appropriate to your application.
Measure cost, latency, and outcomes in production
For each request, log the selected model, input and output usage, latency, task class, and an outcome signal such as a review result or task-specific correctness check. Use those records to compare the policy with alternatives on your actual traffic; task frequency matters because a small per-call difference can accumulate in repeated automation.
Calculate spend from observed usage and the current applicable rates. At the listed standard rates, estimated uncached text-token cost is:
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →cost_usd = (input_tokens / 1_000_000 * input_rate) +
(output_tokens / 1_000_000 * output_rate)For cached input, account for the separately listed cached-input rate where applicable rather than pricing those tokens as ordinary input. This estimate covers only the token categories and rates shown; verify current pricing and any other charges relevant to your API use.
Do not infer a latency winner from the tier labels or prices. Measure response times under your own traffic and request conditions. If you need a different processing service tier, the Responses API reference documents service_tier values fast and priority and says the response reports the tier actually used. Treat that as a separate choice from selecting Sol, Terra, or Luna, and verify current eligibility and pricing before relying on it: Responses API reference.
Check API limits and availability before deployment
As checked October 7, 2026, the three model pages list the same headline context window of 1,050,000 tokens, maximum output of 128,000 tokens, and reasoning-effort choices of none, low, medium, high, xhigh, and max. Confirm current availability for your account, supported tools, and request behavior in the API surface you use; model-page limits and parameters can change. Consult the individual Sol, Terra, and Luna pages at implementation time.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
