STEALTHMARK · HEAD TO HEAD

All comparisons ↗

GPT-6 Luna vs Hy4 preview

Hy4 preview leads GPT-6 Luna by 4.8 points on StealthMark, 43.9 to 39.1. It leads on 5 of 9 benchmarks, including Foresight, and GPT-6 Luna on 4, including Latch. Hy4 preview scores higher on Newton's Mirror, with its largest category advantage in reflections.

See the difference ↘
All results ↗
All results ↗

Overall StealthMark

39.1/100Overall rank 40/58
43.9/100Overall rank 34/58

General Intelligence

31.3/100
41.3/100

Visual Intelligence

52.0/100
48.6/100

Reasoning efficiency

How many reasoning tokens a model needs for the quality it delivers.

Single-step reasoning How much it reasons to understand a task

Multi-step reasoning How much it keeps reasoning as a task goes on

Cost efficiency

How much Performance you get for your money.

Open or local

Open modelApache-2.0

BENCHMARK BY BENCHMARK

Where each model leads

  • GPT-6 Luna
  • Hy4 preview
  • Other models
  • Median

DATASHEET

Specs and pricing

GPT-6 Luna is 5× cheaper per output token and runs 2.1× faster in our speed test. Hy4 preview scores higher on StealthMark. Both take about 1.05M tokens of context.

  • GPT-6 Luna
  • Hy4 preview
  • Other models
  • Median
PricePer 1M output tokens, input below
GPT-6 Luna: $0.505× cheaper$0.10 inputHy4 preview: $2.50$0.83 input
Context windowTokens per request, max output below
GPT-6 Luna: 1.05MTied128K max outputHy4 preview: 1MTied64K max output
Output speedTokens per second, our measurement
GPT-6 Luna: 1312.1× fastertokens/sHy4 preview: 64tokens/s
SizeTotal and active parameters
GPT-6 Luna: Not disclosedHy4 preview: 770B49B active
Developer
OpenAITencent Hy Team
Released
22 Sep 202628 Aug 2026
Origin
United StatesChina
Input → output
Text, image → textText → text
Architecture
Not publishedMoE; 78-layer backbone with Gated DSA, 256 routed experts + 1 shared expert, top-8 routed experts active, plus native MTP layer

Full datasheets, every price and sourceGPT-6 Luna ↗Hy4 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-6 Luna 29.8/100Rank 37/54

Survived 0 of 3
Foresight: complete playing field
Level 36 ptsTurn 7 / 20

Hy4 preview 53.1/100Rank 17/54

Survived 1 of 3
Foresight: complete playing field
Level 311 ptsTurn 16 / 20

Points per level

LevelGPT-6 LunaHy4 preview
Level 1Second Thoughts
15 ptsOut of lives
26 ptsSurvived
Level 2Pressure Pot
5 ptsOut of lives
6 ptsOut of lives
Level 3Hide or Flight
6 ptsOut of lives
11 ptsOut of lives

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-6 Luna 44.0/100Rank 31/55

Cleared 4 of 5
Latch: complete playing field
Level 513 ptsTurn 15 / 40

Hy4 preview 35.6/100Rank 38/55

Cleared 3 of 5
Latch: complete playing field
Level 518.2 ptsTurn 19 / 40

Points per level

LevelGPT-6 LunaHy4 preview
Level 1Hold the door
5.2 ptsCleared
5.2 ptsCleared
Level 2Crossfeed
8.8 ptsClearedFinished early · 1 beacon left
6 ptsFinished early
Level 3Relay vault
17 ptsCleared1 beacon left
18 ptsCleared
Level 4Passing bay
15.8 ptsClearedFinished early · 3 beacons left
15.8 ptsClearedFinished early · 3 beacons left
Level 5Second thoughts
13 ptsFinished early
18.2 ptsFinished early

Newton's Mirror

Same task instructions

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

The judgment

GPT-6 Luna

The schematic scene supports the paint contact order and compact paint scale. Its low, undersized main sphere and incorrectly placed wall images depart from the reference geometry. A few shrinking wall images provide limited optical organization, while absent lateral recursion and outer chains, broad sphere ribbons and isolated panel highlights leave most nested room and emitter mapping undeveloped.

Hy4 preview

The central sphere, compact airborne black bead and broad paint ordering are supported. Contact and transport history remain partly unresolved. The optical system provides only isolated first images and light patches, with little evidence of the extensive ordered reflection structure. Rounded metal and paint gloss are more successful than the reflected contents.

Score breakdown

CategoryGPT-6 LunaHy4 preview
State at 600 ms22.0/2518.0/25
Camera and placement3.5/1511.5/15
Reflections5.0/12029.5/120
Materials and scale7.0/107.5/10
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-6 Luna

Sixfold symmetry, pale coloring and six outward-facing heads establish the motif. However, every neck floats clear of the knot. Squared outer sections replace the required semicircular lobes, and the large stepped opening lacks a regular hexagonal contour. Repeated scales and soft shading suggest skin, but translucent crossings and doubled edges weaken the living, interwoven construction.

Hy4 preview

A cohesive, appealing illustration with a serene face, luminous palette and complete framing. The separated head and torso weaken anatomical coherence, while crowded wing attachments and angular sash folds limit the requested suppleness. Clean execution and decorative detail exceed the clarity of the underlying construction.

Score breakdown

CategoryGPT-6 LunaHy4 preview
Task & likeness4.5/106.0/10
Anatomy & construction3.5/105.5/10
Space & placement5.5/107.5/10
Light & reflections5.0/107.0/10
Purposeful detail5.0/106.5/10
Artistry & character5.0/105.0/10
Materials & effects5.0/107.0/10
Visual integrity4.5/105.5/10

Benchmark task

GPT-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.

Hy4 preview

A young blue dragon floating among clouds tests curved anatomy, expressive features and the contrast between soft clouds and glossy surfaces.

Mona Lisa

Different task instructions

Precise visual recall and portrait reconstruction.

The judgment

GPT-6 Luna

The dark veil, crossed arms, restrained smile and layered landscape clearly evoke Mona Lisa. Clean contours and muted colors create a coherent illustration. Replication is limited by enlarged, outlined eyes, simplified facial volume, broad shoulders and indistinct overlapping hands. Gradients suggest depth, but lack the original's subtle painterly transitions.

Hy4 preview

A recognizable but basic Mona Lisa interpretation with a cohesive earthy palette and atmospheric hills. The rigid neck, generic face and unresolved folded-hand pose limit replication and anatomy. Large silhouettes and layer boundaries are comparatively clear, warranting stronger integrity credit than the sparse detail and simplified construction.

Score breakdown

CategoryGPT-6 LunaHy4 preview
Task & likeness4.5/102.5/10
Anatomy & construction4.0/102.5/10
Space & placement5.5/102.5/10
Light & reflections5.0/101.5/10
Purposeful detail5.0/102.5/10
Artistry & character4.5/101.5/10
Materials & effects5.0/102.5/10
Visual integrity6.0/104.0/10

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-6 Luna

The warm sky, layered sea and framing palms create an appealing beach composition. Animals, birds, toys and reflected sunlight fulfill the brief. Human anatomy remains schematic, and simplified figures offer limited expression. Missing shadows and the dolphin's splashless jump weaken physical grounding, while clean silhouettes and readable small details support the illustration.

Hy4 preview

A cheerful beach illustration fulfills the visible subject requirements with coherent foreground construction and a harmonious sunset palette. Shallow depth, schematic reflections, ambiguous blue airborne forms and exposed geometry beneath the sunset limit its finish. Small accessories are readable, but materials and surface detail remain simplified.

Score breakdown

CategoryGPT-6 LunaHy4 preview
Task & likeness7.5/108.5/10
Anatomy & construction5.0/105.5/10
Space & placement7.5/106.5/10
Light & reflections5.0/105.0/10
Purposeful detail7.0/106.0/10
Artistry & character5.0/105.5/10
Materials & effects6.5/105.0/10
Visual integrity6.0/106.0/10

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-6 Luna

This benchmark has not been assessed yet.

Hy4 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-6 Luna

Expressive faces, warm sunset colors and clean silhouettes create an appealing illustration. The raccoon sits on the dog's back, but the dog's grip and steering linkage remain unclear. The broad hull resembles a speedboat, and flat shading, absent sun reflections and modest water displacement limit depth and motion.

Hy4 preview

A lively cartoon communicates the animal pair traveling across water. Curving waves, rising spray and locally modeled gray forms strengthen motion and volume. Ambiguous seating, simplified joints and unclear control routing limit construction. Competing focal elements, decorative foam and inconsistent illumination leave the finish uneven.

Score breakdown

CategoryGPT-6 LunaHy4 preview
Task & likeness8.0/107.0/10
Anatomy & construction5.5/104.0/10
Space & placement7.5/106.0/10
Light & reflections6.0/106.5/10
Purposeful detail7.0/105.5/10
Artistry & character5.5/104.5/10
Materials & effects6.5/105.5/10
Visual integrity5.0/105.0/10

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-6 Luna

The pelican’s distinctive bill and the coordinated vintage poster palette make an appealing illustration. Circular wheels, orderly spokes and clean lettering strengthen the design. Riding mechanics are less convincing: neither wing reaches the handlebars, pedal contact is ambiguous, and the drivetrain is incomplete. The wheels sit against water without a clear supporting surface. Flat lighting limits volume.

Hy4 preview

A recognizable, playful pelican cyclist with readable wheels, accessories and landscape layers. Form shading and ground shadows provide competent depth. Angular legs, uncertain pedal engagement and absent steering contact limit the riding construction; crowded overlaps and simple surfaces constrain refinement.

Score breakdown

CategoryGPT-6 LunaHy4 preview
Task & likeness8.0/107.5/10
Anatomy & construction5.0/104.5/10
Space & placement6.5/106.5/10
Light & reflections6.5/107.0/10
Purposeful detail7.5/106.0/10
Artistry & character6.5/105.0/10
Materials & effects6.5/105.0/10
Visual integrity7.0/107.0/10

Benchmark task

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