On November 10, 2026, in the Claudexia Premium channel, Luna costs $0.04 / $0.20 per million input / output tokens, and the GPT-5.6 lineup has two middle tiers above it: gpt-5.6-sol at $0.24 / $1.16 and gpt-5.6-terra at $0.34 / $1.62. Both cover ground where Luna isn't enough and Opus 5 ($1.08 / $4.12) is overkill.
Terra: a step up from Luna
Terra costs about 8.5 times what Luna does on input and about 8 times on output, and reasons noticeably more confidently. Reach for it when the task is a notch past pure classification: explain a decision in a couple of sentences, reconcile data from a few sources, handle mixed input that doesn't fit a rigid template. It's a mid-pipeline model: accuracy matters, but you don't need multi-step logic.
Sol: the top GPT tier
Sol costs $0.24 / $1.16, less than claude-sonnet-5 at $0.52 / $1.98. Price-wise it's a direct competitor to the Claude mid-tier. Reach for Sol when you specifically want a GPT-family model: a different response style, different behavior on certain prompts, or compatibility with an existing pipeline already built around GPT.
Terra versus Sol: how to choose
Terra costs about 1.4 times what Sol does ($0.34 / $1.62 against $0.24 / $1.16), so the simple rule runs the other way: run a test set through both and compare errors, not general impressions of the output. If Sol's answers are as good, take the cheaper one. If Terra keeps winning on edge cases, the gap of $0.10 per million input tokens and $0.46 per million output tokens pays for itself quickly in fewer retries.
Sol versus Sonnet: the price gap
Sol costs $0.24 / $1.16 per million input / output tokens and claude-sonnet-5 costs $0.52 / $1.98, so Sonnet runs about 2.2 times as expensive on input and 1.7 times on output. If your pipeline is already tuned around Claude, Sonnet's output style will feel more familiar, and the price gap is what that familiarity costs. If you're keeping both families for resilience against one provider's outages, Sol is a solid and cheaper backup.
A working example
from anthropic import Anthropic
client = Anthropic(
api_key="sk_cdx_your_key_here",
base_url="https://api.claudexia.tech",
)
message = client.messages.create(
model="gpt-5.6-sol",
max_tokens=1024,
messages=[{"role": "user", "content": "Merge these three reports into one table and flag the discrepancies."}],
)
print(message.content[0].text)
The same call over OpenAI Chat Completions differs only in the request body shape — the key and the base URL stay the same.
Mixing models in one pipeline
Nothing stops you from running Luna on a preprocessing step and Sol on the step that needs a sharper answer. One key works across the whole lineup, both API formats live on one endpoint, so switching a model is a one-line config change, not a new integration.
For example: Luna filters the incoming stream first and drops obvious noise, then Sol works through what's left and writes the full answer. High volume and price sensitivity on the first step, lower volume and accuracy on the second.
Comparing on your own task
The calculator on the site works out the bill difference between Terra, Sol, and Sonnet at your volumes. Per-key stats show which model is actually spending your budget in production, not in a test run.
In short
Terra for mid-pipeline steps that need accuracy without deep reasoning. Sol when you want a GPT-family model priced below Sonnet. Between Sol and Sonnet the price gap is real ($0.24 / $1.16 against $0.52 / $1.98 on November 10, 2026), so test both on your own task.