GPT-5.6 Luna API — High-Volume Cheap LLM

GPT-5.6 Luna is KeyoAPI’s high-volume GPT-class chat model for cheap bulk LLM work on an OpenAI-compatible endpoint.

Listed price: ~$0.10 / $0.62 per 1M in/out · Confirm live rates on interactive pricing.

Overview

GPT-5.6 Luna is KeyoAPI's high-volume GPT-class chat model for workloads where unit cost matters more than peak model IQ. You keep an OpenAI-compatible client (base URL https://www.keyoapi.xyz/v1, Bearer API key) and set model=gpt-5.6-luna on POST /v1/chat/completions. Indicative sell rates are about $0.10 input / $0.62 output per 1M tokens — confirm live on /pricing/gpt-5.6-luna before you budget.

Who should use Luna. Put Luna on the paths that burn most tokens: ticket tagging, email triage, log classification, SEO or localization drafts, JSON tool-result summarization, and nightly batch rewrites. Keep product chat defaults on gpt-5.6-terra (or Claude / Sol) and escalate only when automated quality gates fail. That split is how lower token rates show up in real cost of goods instead of living only in a pricing spreadsheet.

How Luna differs from Terra and Sol. Luna trades some peak intelligence for aggressive unit cost. Terra (~$0.37 / $2.21 per 1M in/out indicative) is the everyday GPT-class default. Sol (~$0.92 / $5.52) is the denser escalation tier. Guides: /model/gpt-5.6-terra and /model/gpt-5.6-sol. Side-by-side OpenAI list context: /openai-api-pricing. Catalog compare: /compare.

Worked planning example (indicative). Suppose a classifier burns 80M input tokens and 8M output tokens per month. At Luna rates that is roughly $8.00 input + $4.96 output ≈ $12.96 before retries. Moving the same mix to Terra is roughly $29.60 + $17.68 ≈ $47.28. If 10% of rows fail a JSON-schema check and escalate to Terra, you still keep most spend on Luna. Recompute with live /pricing figures and your real retry rate.

Integration steps. (1) Sign up at /sign-up and top up Wallet. (2) Create an API key. (3) Point your OpenAI SDK or HTTP client at https://www.keyoapi.xyz/v1. (4) Call chat completions with model gpt-5.6-luna. (5) Try/buy from the interactive Model Square page /pricing/gpt-5.6-luna — there is no separate upload playground. (6) Log usage.prompt_tokens and usage.completion_tokens from each response so finance sees actuals, not guesses.

Quality and routing pattern. Validate Luna outputs with cheap checks first: JSON schema, required fields, banned phrases, length caps. Route only failing slices to Terra or Sol. Cache identical prompts with a hash key when the job is deterministic. Truncate unused chat history on batch jobs. Prefer short structured outputs for classification so you are not paying for prose you discard.

When not to use Luna alone. Long multi-step reasoning, high-stakes customer replies, and dense coding agents usually need Terra, Sol, or Claude (/claude-api-pricing). Luna is the volume layer under those models, not a replacement for every call.

Free prototyping on the same key. Keyo does not ship an OpenAI-branded free GPT id. Use permanent $0 twins such as deepseek-v4-flash-free or glm-5.2-free from /free-models to wire the client, then flip model= to gpt-5.6-luna when quality gates pass — same base URL and auth.

Operational checklist. Confirm live rates after catalog changes; set timeouts and retries with jitter; cap max_tokens on batch; store request ids for support; and keep a feature flag so you can shift traffic to Terra during incidents without redeploying your whole stack. Fund prepaid credits before load tests so 402s do not look like model failures.

Luna exists because most production tokens are boring labels, rewrites, triage, and short transforms. Those jobs should never hit flagship rates by default. Queue Luna, batch when possible, escalate the failing fraction, and measure cost from usage objects — that is the practical cheap LLM API pattern on KeyoAPI.

Tier cheat-sheet (indicative Keyo sell rates — confirm live). Luna ~$0.10 / $0.62 per 1M in/out for volume. Terra ~$0.37 / $2.21 for everyday product chat. Sol ~$0.92 / $5.52 when you need denser reasoning. OpenAI list bands for GPT-4o mini / GPT-4o style planning context sit on /openai-api-pricing; always verify both Keyo and OpenAI public pages before you sign a volume plan.

Copy-paste client sketch. Export KEYO_API_KEY, set OPENAI_BASE_URL=https://www.keyoapi.xyz/v1, then either curl chat/completions or use the official OpenAI Python/Node SDK with base_url pointed at Keyo. Pass model=gpt-5.6-luna for the cheap path. After each response, persist usage.prompt_tokens and usage.completion_tokens next to your job id so month-end finance matches the Wallet ledger.

Failure modes to plan for. Empty Wallet returns payment errors — top up before load tests. Wrong model strings fail loudly; copy ids from /pricing-list or Model Square. Over-long contexts inflate cost without improving classification; truncate. If Luna output fails schema validation repeatedly on a niche domain, move that niche to Terra permanently instead of paying double latency with endless retries.

Related reading on the same site. Model Square try/buy: /pricing/gpt-5.6-luna. Sibling guides: /model/gpt-5.6-terra, /model/gpt-5.6-sol. Vendor landing: /openai-api-pricing. Free on-ramp: /free-models. Docs: /brand/keyo-docs.html. FAQ: /brand/faq.html.

Quick start

# cURL
curl https://www.keyoapi.xyz/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.6-luna","messages":[{"role":"user","content":"Tag this ticket: refund delayed shipping"}]}'

# Python (OpenAI SDK)
from openai import OpenAI
client = OpenAI(base_url="https://www.keyoapi.xyz/v1", api_key="YOUR_KEYO_API_KEY")
r = client.chat.completions.create(
    model="gpt-5.6-luna",
    messages=[{"role": "user", "content": "Return JSON {\"label\": \"refund|shipping|other\"} for: order late"}],
)
print(r.choices[0].message.content)
print("tokens", r.usage)

FAQ

Related models

Guide for gpt-5.6-luna. Try/buy: /pricing/gpt-5.6-luna · Catalog: /pricing.