STEALTHMARK · HEAD TO HEAD

All comparisons ↗

GPT-5.6 Luna vs Step 5 Preview

Step 5 Preview leads GPT-5.6 Luna by 1.2 points on StealthMark, 39.0 to 37.8. GPT-5.6 Luna leads on 7 of 9 benchmarks, including both games, and Step 5 Preview on 2. Step 5 Preview scores higher on Newton's Mirror, with its largest category advantage in reflections.

See the difference ↘
All results ↗
All results ↗

Overall StealthMark

37.8/100Overall rank 37/53
39.0/100Overall rank 36/53

General Intelligence

30.4/100
37.1/100

Visual Intelligence

50.2/100
41.9/100

Reasoning efficiency

On every task, result quality and reasoning tokens are each compared with all measured models and weighed equally. Each dot is one model.

Single-step reasoning Result quality weighed against reasoning tokens on a single problem

Multi-step reasoning Result quality weighed against reasoning tokens across connected steps, as in agentic workflows

  • GPT-5.6 Luna

BENCHMARK BY BENCHMARK

Where each model leads

  • GPT-5.6 Luna
  • Step 5 Preview
  • Other models
  • Median

DATASHEET

Specs and pricing

Both take about 1.05M tokens of context.

  • GPT-5.6 Luna
  • Step 5 Preview
  • Other models
  • Median
PricePer 1M output tokens, input below
GPT-5.6 Luna: $1.20$0.20 inputStep 5 Preview: No list price
Context windowTokens per request, max output below
GPT-5.6 Luna: 1.05MTied128K max outputStep 5 Preview: 1MTied
Output speedTokens per second, our measurement
GPT-5.6 Luna: 181tokens/sStep 5 Preview: Not measured
SizeTotal and active parameters
GPT-5.6 Luna: Not disclosedStep 5 Preview: 600B27B active
Developer
OpenAIStepFun
Released
9 Jul 2026Not published
Origin
United StatesNot published
Input → output
Text, image → textText, image, video → text
Architecture
Not publishedSparse mixture-of-experts; 92 layers (88 MoE), 352 routed experts with 8 selected per token plus a shared expert; grouped-query attention with 512-token sliding windows and full attention every fourth layer; 3 multi-token-prediction layers; vision tower for image and video input

Full datasheets, every price and sourceGPT-5.6 Luna ↗Step 5 Preview ↗

Anticipating threats and planning ahead. Collect coins while avoiding ghosts. Ghosts move before the player each turn, and collisions cost a life. 3 levels in sequence, up to 66 turns in total. Earlier moves and outcomes carry forward, testing learning across levels.

GPT-5.6 Luna 25.1/100Rank 36/49

Survived 0 of 3
Foresight: complete playing field
Level 38 ptsTurn 15 / 20

Step 5 Preview 59.9/100Rank 42/491 of 3 levels

1 of 3 played
Foresight: complete playing field
Level 122 ptsTurn 28 / 28

Points per level

LevelGPT-5.6 LunaStep 5 Preview
Level 1Second Thoughts
4 ptsOut of lives
22 ptsSurvived
Level 2Pressure Pot
6 ptsOut of lives
Pending
Level 3Hide or Flight
8 ptsOut of lives
Pending

Cooperative planning and learning from experience. Coordinate two robots to push numbered cores onto matching sockets. Pressure plates hold doors open; moving either robot uses one turn. 5 levels in sequence, up to 152 turns in total. Earlier moves and outcomes carry forward, testing learning across levels.

GPT-5.6 Luna 52.9/100Rank 23/50

Cleared 5 of 5
Latch: complete playing field
Level 520.6 ptsTurn 27 / 40

Step 5 Preview 40.7/100Rank 28/50

Cleared 4 of 5
Latch: complete playing field
Level 515 ptsTurn 35 / 40

Points per level

LevelGPT-5.6 LunaStep 5 Preview
Level 1Hold the door
5.2 ptsCleared
5.2 ptsCleared
Level 2Crossfeed
8.6 ptsClearedFinished early · 1 beacon left
9 ptsCleared
Level 3Relay vault
18 ptsCleared
18.6 ptsCleared
Level 4Passing bay
17 ptsClearedFinished early · 3 beacons left
15.6 ptsClearedFinished early · 3 beacons left
Level 5Second thoughts
20.6 ptsCleared
15 ptsFinished early

Newton's Mirror

Same task instructions

Physics, perspective and recursive reflections, tested together. Explore the test ↗

The judgment

GPT-5.6 Luna

The room enclosure has coherent symmetric perspective, with the overhead panel, sphere and black aligned centrally. The sphere and airborne black occupy substantially lower positions than the reference framing, inconsistent with the prescribed camera aim. The diagonal crimson streak does not establish the required surface transport. Faint wall sphere images provide limited mirrored correspondence, but the shallow floor image, missing receding chains, absent nested sphere imagery and sparse emitter mapping leave most reflection structure unsupported.

Step 5 Preview

The main sphere, airborne black and small colored features occupy broadly appropriate positions. Most of the defining mirror system is absent, and paint interaction remains only partially supported. The small paint features do not visibly establish a separate volume-scale violation.

Score breakdown

CategoryGPT-5.6 LunaStep 5 Preview
State at 600 ms15.0/2520.0/25
Camera and placement4.0/1513.5/15
Reflections3.5/12020.0/120
Materials and scale6.5/107.0/10
Floor-damage reward0.0/100.0/10

Benchmark task

A spinning metal sphere and paint in a mirrored room test physical reasoning, perspective and consistency across repeated reflections.

Selfie

Different task instructions

Visual construction and creative interpretation.

The judgment

GPT-5.6 Luna

Six pale, outward-facing heads and six rounded lobes make the knot recognizable. The central hexagonal opening remains clear. However, broad gray crossing panels flatten the weave, several inner joins resemble capped strips, and sparse oversized scale marks leave the bodies looking more like decorated ribbons than living snakes.

Step 5 Preview

A charming diagonal climber and warm palette communicate the central idea. Simple rounded construction, crowded limb counting, uncertain footholds and an unattached upper shelf limit its physical coherence. Growth bands and the yarn trail add specificity, but the requested material distinctions and unravelled bootie remain underdeveloped.

Score breakdown

CategoryGPT-5.6 LunaStep 5 Preview
Task & likeness7.5/106.0/10
Anatomy & construction6.0/104.5/10
Space & placement6.5/105.0/10
Light & reflections5.0/105.5/10
Purposeful detail5.5/105.5/10
Artistry & character6.0/105.0/10
Materials & effects5.0/106.0/10
Visual integrity6.0/105.0/10

Benchmark task

GPT-5.6 Luna

A pale serpent woven into an interlocking knot tests connected anatomy, repeated structure and readable overlaps. Organic surface detail must work with the overall geometry.

Step 5 Preview

Character scenes test expression, anatomy, materials and coherent visual construction.

Mona Lisa

Same task instructions

Precise visual recall and portrait reconstruction.

The judgment

GPT-5.6 Luna

The central portrait, dark clothing, crossed hands and muted landscape clearly evoke Mona Lisa. Replication is limited by the elongated face, conspicuous eyes and simplified smile. Overlapping fingers have uncertain anatomy, while broad gradients and heavy outlines flatten the portrait. The composition and restrained palette remain coherent.

Step 5 Preview

A recognizable, coherent simplification with purposeful facial contours and differentiated fingers. These provide more construction and detail than the broad silhouette alone suggests. The oversized face, frontal seated pose, unresolved wrists and sparse landscape still substantially limit likeness, depth and finish.

Score breakdown

CategoryGPT-5.6 LunaStep 5 Preview
Task & likeness3.5/104.0/10
Anatomy & construction3.5/104.0/10
Space & placement4.0/104.0/10
Light & reflections4.5/103.5/10
Purposeful detail4.0/104.0/10
Artistry & character4.0/104.0/10
Materials & effects3.5/103.0/10
Visual integrity4.5/103.5/10

Benchmark task

An SVG interpretation of the Mona Lisa tests facial likeness, anatomy and the composition of a familiar portrait.

Beach

Same task instructions

Creativity, perspective and visual beauty.

The judgment

GPT-5.6 Luna

The warm sunset, palm silhouette and varied beach activities create an appealing, readable composition. Animals, birds and a broad sun reflection satisfy the brief. Drawing remains elementary: angular airborne sea animals lack splash or convincing motion, the crab's claws are awkwardly constructed, and repeated reflection bands suggest a stepped platform. Clean outlines and pleasant colors exceed the anatomical and spatial sophistication.

Step 5 Preview

A recognizable sunset beach with clear warm-cool contrast and playful accessories. Basic animal construction, shallow foreground placement and schematic reflections limit spatial and physical credibility. Sparse descriptive detail and abruptly ended sky bands constrain the finish.

Score breakdown

CategoryGPT-5.6 LunaStep 5 Preview
Task & likeness7.5/104.0/10
Anatomy & construction4.0/103.0/10
Space & placement5.5/103.5/10
Light & reflections4.0/103.5/10
Purposeful detail6.5/103.0/10
Artistry & character3.5/103.0/10
Materials & effects4.0/103.5/10
Visual integrity3.5/104.0/10

Benchmark task

A busy shoreline at sunset tests scene composition, animal forms and the relationship between light and water reflections.

Ground Truth

Same task instructions

Physical reasoning, light and visual reconstruction.

The judgment

GPT-5.6 Luna

This benchmark has not been assessed yet.

Step 5 Preview

This benchmark has not been assessed yet.

Benchmark task

A plank, computer mouse and blueberry in a steel room test physical stability, spatial relationships, materials and lighting.

Raccoon

Same task instructions

Scene composition and interacting subjects.

The judgment

GPT-5.6 Luna

Expressive animals, a striped tail and clean outlines make the scene appealing and readable. The raccoon sits convincingly on the dog's back, but the overlapping steering paws are ambiguous and the control stem stops above the hull. Broad wake shapes suggest motion; the glowing sun lacks a corresponding water reflection.

Step 5 Preview

Recognizable animals, a bright palette and identifying accessories support a basic cartoon. Primitive anatomy, partial seated support, unresolved steering and weak hull-water contact limit the central action. Lighting and spray provide atmosphere without establishing convincing physical depth.

Score breakdown

CategoryGPT-5.6 LunaStep 5 Preview
Task & likeness7.0/104.0/10
Anatomy & construction5.0/104.5/10
Space & placement6.5/103.0/10
Light & reflections6.0/103.5/10
Purposeful detail6.5/104.0/10
Artistry & character4.5/103.5/10
Materials & effects5.5/103.0/10
Visual integrity5.0/103.0/10

Benchmark task

A jet-ski scene pairs a canine driver with a raccoon passenger. It tests the placement and contact of interacting subjects, along with vehicle construction.

Pelican

Same task instructions

Subject construction and spatial relationships.

The judgment

GPT-5.6 Luna

A recognizable pelican cyclist with an expressive profile, balanced composition and controlled outlines. Head, neck and torso connect coherently. The bicycle wheels and frame are convincing, but overlapping feet obscure pedal contact, the drivetrain lacks a clear chain, and the wing's handlebar grip is schematic. Atmospheric color succeeds more strongly than directional lighting or material depth.

Step 5 Preview

The pelican and bicycle are recognizable, with a pleasant palette and basic diffuse shading. The riding relationship fails physically: the bird is unsupported, the front wheel is unmounted, and pedalling is unresolved. Simple geometric construction and sparse material treatment further limit the drawing.

Score breakdown

CategoryGPT-5.6 LunaStep 5 Preview
Task & likeness8.5/103.0/10
Anatomy & construction5.5/103.0/10
Space & placement7.5/103.0/10
Light & reflections6.5/103.0/10
Purposeful detail6.5/104.0/10
Artistry & character5.5/103.0/10
Materials & effects7.0/103.0/10
Visual integrity6.5/102.0/10

Benchmark task

A cycling pelican tests anatomy, riding posture and contact between the bird and the bicycle.