High-throughput, low-latency tier at about one-fifth of Sol’s price
Luna targets voice agents, chat, and real-time automation at high request volume.
Overview
GPT-5.6 Luna is the fastest, most cost-efficient tier in OpenAI's GPT-5.6 family, designed for low-latency, high-throughput workloads such as voice agents, chatbots, and real-time automation. It launched alongside Sol and Terra — limited preview June 26, 2026, broad API access July 9, 2026 — as the entry point into the GPT-5.6 lineup.
Luna keeps the same 1.05M-token context window and 128K max output as its siblings, at roughly one-fifth of Sol's per-token price, making it the natural default for high-volume requests where per-call cost and speed matter more than peak reasoning depth.
Key capabilities
| Dimension | Detail |
|---|---|
| Context window | 1,050,000 tokens |
| Max output | 128,000 tokens |
| Input modalities | Text, image |
| Output modalities | Text |
| Reasoning effort | none, low, medium, high, xhigh, max (default: medium) |
Inputs above 272K tokens enter a higher long-context tier (2x input, 1.5x output). Cached input reads get the standard ~90% discount. See live pricing in the model catalog.
Benchmarks and boundaries
The cheapest tier, still usable for everyday coding tasks
Source: Artificial Analysis Intelligence Index and Coding Agent Index.
On Artificial Analysis's Intelligence Index (max reasoning effort), Luna scores 51 — behind Terra's 55 and Sol's 59 — at roughly one-fifth of Sol's per-task cost (about $0.21). On the Coding Agent Index inside Codex, Luna reaches 75, still usable for routine coding help even though the hardest, longest-horizon tasks are better routed to Terra or Sol.
When to use it
- High-volume, latency-sensitive workloads: voice agents, chat interfaces, and real-time automation where speed and per-call cost dominate.
- Simple to moderate coding and reasoning tasks: routine edits, short scripts, and classification-style work.
- First-pass routing layer: handle the bulk of easy requests and escalate only the hard cases to Terra or Sol.
CrossModel exposes GPT-5.6 Luna through an OpenAI-compatible /v1/chat/completions API. Current pricing is available in the model catalog.