OpenAI
GPT-5.6 Luna
A delivery fallback the team rarely reaches for while Terra remains accessible.
Canonical model record
Current identity, limits, and pricing
- Status
- Current
- API model ID
gpt-5.6-luna- Context
- 1.05M
- Max output
- 128K
- API price / 1M tokens
- $1 input · $6 output
Visual prompt runs
Benchmark runs
Open each generated scene, or compare the same prompt across models.
Helm's Deep
Fortress siege scene testing scale, lighting, architecture, and cinematic atmosphere.
Hogwarts Broom Flight Simulator
Broom-flight scene testing depth, motion cues, castle scale, and fantasy mood.
Jabberwock
Dark fantasy encounter testing creature design, forest mood, and narrative staging.
Low Poly World
Stylized island build testing composition, color, and low-poly worldbuilding.
Office Life
Workplace vignette testing everyday scene logic, objects, and believable office detail.
Petri Dish
Microscopic ecosystem testing organic forms, scientific clarity, and cellular detail.
Universe Simulator
Cosmic system testing orbital structure, glowing bodies, scale, and simulation readability.
Vice City
Neon coastal city testing vehicles, architecture, atmosphere, and dense urban layout.
Yingzao Fashi Assembly
Timber assembly scene testing structure, joinery, construction order, and material clarity.
Official benchmark profile
How GPT-5.6 Luna scores beyond our visual tests.
OpenAI describes Luna as the fastest and most cost-efficient GPT-5.6 variant. These rows use the shared GPT-5.6 comparison table so scores line up against Sol, Terra, GPT-5.5, Claude, and Gemini where published.
SWE-bench Pro
62.7%Terminal-Bench 2.1
84.7%OSWorld 2.0
45.6%Full official benchmark table9 rows with source settings and peer charts
| Benchmark | Area | Score | Setting / comparison |
|---|---|---|---|
| SWE-bench Pro | Coding | 62.7% | Shared OpenAI GPT-5.6 comparison table. |
| Terminal-Bench 2.1 | Agentic coding | 84.7% | Shared OpenAI GPT-5.6 comparison table. |
| OSWorld 2.0 | Computer use | 45.6% | High-effort computer-use comparison table. |
| BrowseComp | Tool use | 83.3% | High-effort browsing-agent comparison table. |
| BenchCAD | Computer-aided design | 63.1% | Vision2Code score without Python tool. |
| BenchCAD with Python tool | Tool use | 73.9% | Vision2Code score with Python tool. |
| GPQA Diamond | Academic reasoning | 92.3% | Shared reasoning benchmark table. |
| FrontierMath Tier 1-3 v2 | Math | 78.6% | Shared reasoning benchmark table. |
| FrontierMath Tier 4 v2 | Math | 58.5% | Shared reasoning benchmark table. |
Superbash commentary
Our take
GPT-5.6 Luna sits in C tier. The team sees it as a delivery model, but in practice uses Terra on low reasoning for the same boring work and has not found a strong reason to choose Luna first.
Best for
Watch out
Why it is ranked here
Evidence and commentary
Superbash editorial model ranking
Takeaway: GPT-5.6 Luna is currently placed in Tier C.
The August 2026 editorial roster places GPT-5.6 Luna at rank 13.
Open source →Keep learning