GPT-6 Luna API: The Cheapest GPT-6 Model, Pricing and Setup

Short answer

GPT-6 Luna is the smallest model in the GPT-6 line, available on TeamoRouter as gpt-6-luna with a 272,000-token context window. OpenAI's list price is $0.10 per million input tokens and $0.50 per million output — twenty times cheaper than Sol; on TeamoRouter, as of the publish date, about $0.05 / $0.25. It is built for high-volume, simple work: completions, classification, data extraction, draft edits, subagents. Same key and endpoint as Sol and Astra.

Where Luna sits in the GPT-6 line

All three share the same 272k context. Luna isn't cut down on capacity; it is cheaper because the model is smaller — where deep reasoning isn't needed, the result is the same and the bill is twenty times lower.

Tasks Luna handles

• Autocomplete and short edits in the editor.
• Classification, labeling, extracting fields from text into JSON.
• Generating tests, docstrings, commit messages from existing code.
• Summarizing logs, tickets, documentation.
• Subagents in Codex or Claude Code that read files and search rather than make decisions.
• Batch processing: thousands of similar requests where unit cost matters.

When to move up to Sol

Multi-step refactors, unclear architecture, debugging that needs the logic of dozens of files held at once — that's where Sol earns its price. The rule is simple: start on Luna; if you find yourself redoing answers, step up one tier.

Pricing on TeamoRouter

Per 1 million tokens, in USD:

The TeamoRouter figure tracks the upstream list; check the pricing page for the live number — the table shows publish-date values.

In volume terms: pushing one million input tokens — say, classifying 5,000 support tickets — costs about five cents on TeamoRouter. A subagent session that reads 100k tokens of code and returns a 5k-token digest costs roughly $0.006. The same session on Sol is about $0.04; on Astra about $0.22.

Setup

You need a TeamoRouter API key — create one in the console. Luna uses the same endpoints and formats as the other GPT models.

Calling the API

Chat Completions — the format any OpenAI SDK understands:

from openai import OpenAI

client = OpenAI(base_url="https://api.teamorouter.com/v1", api_key="YOUR_KEY")
resp = client.chat.completions.create(
    model="gpt-6-luna",
    messages=[{"role": "user", "content": "Extract company, amount and due date from this email as JSON: ..."}],
    temperature=0,
)
print(resp.choices[0].message.content)

Responses API — the format Codex uses:

curl https://api.teamorouter.com/v1/responses \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "content-type: application/json" \
  -d '{"model": "gpt-6-luna", "input": "Write a docstring for this function: ..."}'

Full parameter reference: API integration.

Luna in Codex CLI

For cheap sessions, point ~/.codex/config.toml at Luna:

model_provider = "teamorouter"
model = "gpt-6-luna"
model_reasoning_effort = "medium"
[model_providers.teamorouter]
name = "TeamoRouter"
base_url = "https://api.teamorouter.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "responses"

Inside a session, /model lets you step up to gpt-6-sol when the task calls for it. Full Codex installation: guide.

Luna in other agents

In OpenCode, Cline, DeepSeek Harness and any agent with an OpenAI-compatible provider, set the TeamoRouter base_url and the model name gpt-6-luna. If the agent supports a separate "fast" or "background" model for subagents and completions, Luna is the natural fit for that slot.

Paying for usage

You top up a balance in the console and pay only for tokens used — no subscription.

• Card — Visa, Mastercard and other major cards via Stripe.
• USDT — choose "Pay with crypto" in the top-up dialog, send USDT to the address shown; credited within one to five minutes.

No minimum: at Luna prices, even $5 on the balance covers tens of millions of tokens.

FAQ

Does Luna have a free tier? No, Luna is billed per token from the first request. The free models on TeamoRouter are separate IDs with a -free suffix; see the rate limits page for the list and daily caps.

How is Luna different from GPT-5.6 Luna? Same positioning, new generation; the IDs differ (gpt-6-luna vs gpt-5.6-luna) and nothing switches automatically.

Can I mix Luna and Sol in one application? Yes — that's the usual pattern: one key, a model field per request. Cheap stages on Luna, the final decision on Sol or Astra.

Does caching work? Yes, under the same rules as the other GPT models: a repeated prompt prefix is billed at the cache-read rate. Details in the billing FAQ.

Are there request limits? Yes, per model — see the rate limits page.

Next steps

• GPT-6 Sol via API — the workhorse tier for everyday code.
• Claude Opus 5.5 via API — Anthropic's flagship, released the same day.
• Create an API key — and run Luna in five minutes.