STEALTHMARK · HEAD TO HEAD

All comparisons ↗

Step 5 Preview vs Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM) leads Step 5 Preview by 23.6 points on StealthMark, 62.6 to 39.0. It leads on 8 of 9 benchmarks, including Foresight, and Step 5 Preview on 1, including Latch. Union Alpha scores higher on Newton's Mirror, with its largest category advantage in reflections.

See the difference ↘
All results ↗
Model guide ↗

Overall StealthMark

39.0/100Overall rank 36/53
62.6/100Overall rank 8/53

General Intelligence

37.1/100
58.7/100

Visual Intelligence

41.9/100
69.6/100

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

  • Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)Not measured

BENCHMARK BY BENCHMARK

Where each model leads

  • Step 5 Preview
  • Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
  • Other models
  • Median

DATASHEET

Specs and pricing

Step 5 Preview takes 4× more context. Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM) scores higher on StealthMark.

  • Step 5 Preview
  • Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
  • Other models
  • Median
Context windowTokens per request, max output below
Step 5 Preview: 1M4× largerPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM): 262K131K max output
Output speedTokens per second, our measurement
Step 5 Preview: Not measuredPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM): 26tokens/s
SizeTotal and active parameters
Step 5 Preview: 600B27B activePareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM): Not disclosed
Developer
StepFunCircuit & Chisel / Unbiased
Released
Not published20 Sep 2026
Origin
Not publishedUnited States
Input → output
Text, image, video → textText, image → text
Architecture
Sparse 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 inputNot published

Full datasheets, every price and sourceStep 5 Preview ↗Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM) ↗

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.

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

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

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM) 78.7/100Rank 1/49

Survived 2 of 3
Foresight: complete playing field
Level 319 ptsTurn 20 / 20

Points per level

LevelStep 5 PreviewPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
Level 1Second Thoughts
22 ptsSurvived
27 ptsSurvived
Level 2Pressure Pot
Pending
12 ptsOut of lives
Level 3Hide or Flight
Pending
19 ptsSurvived

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.

Step 5 Preview 40.7/100Rank 28/50

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

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM) 36.5/100Rank 32/50

Cleared 4 of 5
Latch: complete playing field
Level 519.4 ptsTurn 33 / 40

Points per level

LevelStep 5 PreviewPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
Level 1Hold the door
5.2 ptsCleared
5.2 ptsCleared
Level 2Crossfeed
9 ptsCleared
9.8 ptsCleared
Level 3Relay vault
18.6 ptsCleared
14.2 ptsClearedFinished early · 4 beacons left
Level 4Passing bay
15.6 ptsClearedFinished early · 3 beacons left
9.2 ptsFinished early
Level 5Second thoughts
15 ptsFinished early
19.4 ptsCleared

Newton's Mirror

Same task instructions

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

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM) 52.0/100 Rank 7/53

Union Alpha: Newton's Mirror SVG benchmark

The judgment

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.

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)

Recognizable near-field sphere placement and bilateral convergence provide limited optical structure. Broad gray regions interrupt lateral sightlines and truncate side-floor images. Missing outer chains, absent nested sphere imagery and incomplete planar and curved emitter mappings dominate the reflection system.

Score breakdown

CategoryStep 5 PreviewPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
State at 600 ms20.0/2524.0/25
Camera and placement13.5/1513.5/15
Reflections20.0/12034.5/120
Materials and scale7.0/1010.0/10
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.

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM) 76.5/100 Rank 7/53

Union Alpha: Selfie SVG benchmark

The judgment

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.

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)

A strong futuristic emblem with clear hierarchy, controlled negative space and distinctive typography. Silver and gunmetal layers give the sweeping ribbons convincing graphic depth, while red accents remain restrained. Ambiguous concealed connections and occasionally abrupt highlight bands limit the construction and material realism.

Score breakdown

CategoryStep 5 PreviewPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
Task & likeness6.0/109.0/10
Anatomy & construction4.5/107.5/10
Space & placement5.0/108.5/10
Light & reflections5.5/107.5/10
Purposeful detail5.5/107.0/10
Artistry & character5.0/107.5/10
Materials & effects6.0/107.5/10
Visual integrity5.0/107.0/10

Benchmark task

Step 5 Preview

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

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)

An interwoven metallic emblem tests ribbon geometry, overlapping forms and the appearance of reflective metal against a dark background.

Mona Lisa

Different task instructions

Precise visual recall and portrait reconstruction.

The judgment

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.

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)

The centered portrait, folded hands, dark clothing and hazy landscape clearly evoke the Mona Lisa. Muted colors and soft facial shading create a coherent illustration, but replication remains schematic: the face is broad, eyes outlined, smile simplified and overlapping fingers poorly resolved. Angular landscape shapes and flat fabric limit depth and painterly richness.

Score breakdown

CategoryStep 5 PreviewPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
Task & likeness4.0/106.5/10
Anatomy & construction4.0/105.0/10
Space & placement4.0/106.5/10
Light & reflections3.5/106.0/10
Purposeful detail4.0/105.5/10
Artistry & character4.0/106.5/10
Materials & effects3.0/105.0/10
Visual integrity3.5/107.0/10

Benchmark task

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

Beach

Different task instructions

Creativity, perspective and visual beauty.

The judgment

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.

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)

A cohesive sunset palette, sweeping shoreline and playful animals make an appealing beach illustration. The sun’s reflection establishes a clear focal axis, balanced by the palm and umbrella. Clean silhouettes and small decorative features reward inspection, though simplified anatomy, repetitive reflection bars and generic oval shadows limit volume and naturalism.

Score breakdown

CategoryStep 5 PreviewPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
Task & likeness4.0/109.0/10
Anatomy & construction3.0/107.0/10
Space & placement3.5/108.0/10
Light & reflections3.5/107.0/10
Purposeful detail3.0/107.0/10
Artistry & character3.0/106.0/10
Materials & effects3.5/106.5/10
Visual integrity4.0/107.0/10

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.

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM) 63.3/100 Rank 8/53

Union Alpha: Ground Truth SVG benchmark

The judgment

Step 5 Preview

This benchmark has not been assessed yet.

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)

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

Different task instructions

Scene composition and interacting subjects.

The judgment

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.

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)

Expressive animals, a readable riding arrangement and sweeping spray create an appealing cartoon. The raccoon sits convincingly on the dog's back, but the dog's paw does not clearly grip the handlebar, whose connection to the hull is unresolved. Flat lighting provides little volume or reflection. A dark rectangular strip interrupts the upper-right background.

Score breakdown

CategoryStep 5 PreviewPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
Task & likeness4.0/108.0/10
Anatomy & construction4.5/106.5/10
Space & placement3.0/107.5/10
Light & reflections3.5/105.5/10
Purposeful detail4.0/107.0/10
Artistry & character3.5/107.0/10
Materials & effects3.0/107.5/10
Visual integrity3.0/107.0/10

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

Different task instructions

Subject construction and spatial relationships.

The judgment

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.

Pareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)

The pelican is immediately recognizable, with an expressive face, convincing bill silhouette and appealing pastel composition. Both legs connect to feet, and the wing reaches the handlebar. Bicycle construction is less resolved: the far pedal lacks a clear crank connection and no chain is visible. Clean contours and orderly spokes support the illustration, while flat lighting and simplified surfaces limit depth.

Score breakdown

CategoryStep 5 PreviewPareto 26.9 (Union of GPT-6 Astra + DeepSeek + GLM)
Task & likeness3.0/108.0/10
Anatomy & construction3.0/106.0/10
Space & placement3.0/107.0/10
Light & reflections3.0/106.0/10
Purposeful detail4.0/106.0/10
Artistry & character3.0/106.0/10
Materials & effects3.0/106.0/10
Visual integrity2.0/108.5/10

Benchmark task

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