Five products. Five identical Unreal tasks. NeoStack completed the most work with the fewest requests and the fewest recorded failures.
Controlled pilot run on July 31, 2026 · One run per product and task · Directional evidence, not a statistical market ranking
The simplest signals of whether an agent can finish real Unreal work efficiently and reliably.
How many of the five Unreal tasks passed every check.
NeoStack · 3× the next-best result
How much back-and-forth the agents needed to do the work.
NeoStack · 9% fewer than next-best Aura
Harness requests that returned a recorded failure.
NeoStack · fewest recorded failures
| Product | Tasks passed | Total requests | Median / task | Total time |
|---|---|---|---|---|
NeoStack AI | 3 / 5 | 145 | 27 | 36.4 min |
Ultimate Engine Copilot | 0 / 5 | 448 | 87 | 58.6 min |
Epic Unreal MCP | 1 / 5 | 375 | 66 | 48.3 min |
Aura | 1 / 5 | 159 | 30 | 61.1 min |
Nwiro Integration Kit | 0 / 5 | 320 | 64 | 34.7 min |
Nwiro’s lower raw elapsed time came from several arms ending early without a strict pass; it completed 0 of 5 tasks.
Pass or fail, request count, elapsed time, and the short reason behind the verdict.
01
A live two-layer warm-core / cool-halo beacon.
NeoStack
50 req · 13.3 min
UE Copilot
101 req · 14.6 min
Epic MCP
103 req · 7.6 min
Aura
23 req · 14.4 min
Nwiro
36 req · 4.7 min
What decided it: NeoStack was the only product to prove both live particle layers and their runtime counts.
02
An animated charcoal, cyan, and orange pulse-grid surface.
NeoStack
27 req · 6.3 min
UE Copilot
111 req · 15.1 min
Epic MCP
47 req · 7.7 min
Aura
36 req · 9.4 min
Nwiro
64 req · 7.6 min
What decided it: Aura was the only product to prove the saved graph, assignment, compile, and same-camera animation.
03
A saved, lit, symmetric signal courtyard.
NeoStack
17 req · 4.1 min
UE Copilot
78 req · 9.9 min
Epic MCP
66 req · 11.4 min
Aura
21 req · 13.0 min
Nwiro
81 req · 8.0 min
What decided it: NeoStack and Epic both proved the saved layout, validation, lighting, symmetry, and required views.
04
A styled signal-relay panel with an opacity animation.
NeoStack
23 req · 6.7 min
UE Copilot
71 req · 9.7 min
Epic MCP
120 req · 15.0 min
Aura
30 req · 8.7 min
Nwiro
63 req · 4.0 min
What decided it: Every product created partial widget work, but none proved the complete saved animation and visual result.
05
A saved Manny locomotion Blend Space and two-state Animation Blueprint.
NeoStack
28 req · 5.9 min
UE Copilot
87 req · 9.4 min
Epic MCP
39 req · 6.7 min
Aura
49 req · 15.6 min
Nwiro
76 req · 10.3 min
What decided it: NeoStack alone proved the Blend Space, state machine, transitions, compilation, and saved result.
Enough methodology to trust the numbers — without turning this page into a lab report.
Same five tasks. Every product received the same held-out Niagara, Material, Level, UMG, and Animation Blueprint work.
Same limits. Each task had 15 minutes and a maximum of 45 state-changing requests.
Same model. Every run used the same model and reasoning configuration in Unreal Engine 5.8.
Strict verification. A pass required saved state, validation, runtime proof when needed, and visual review — not the agent’s own claim.
This pilot used one run per product and task. A strict proof failure can coexist with good-looking or partially correct work; only fully verified outcomes passed.
The Unreal-native agent, available from editor, desktop, web, and mobile.