Swipe for more top models
<report>
Runway
60.53
Megaton Index
The 0–100 weighted total across 15 v-benchmark v2 dimensions, calculated from each published category and technical score at its locked weight.
View methodology →/ 100
Prompt Adherence20%76.03C02Scene Consistency10%78.13C03Physics20%38.82C04Human Fidelity10%67.41C052D Animation Craft & Artistry7.5%47.50C063D Animation Craft & Artistry7.5%44.00C07Cinematography Craft & Artistry2.5%70.00C08Taste & Art Direction6.5%49.66C09Animal & Creature Fidelity1.5%61.43C10Object & Product Fidelity1.5%74.40C11Causal & Semantic Coherence7.5%72.73C12Text Fidelity1.5%47.22C13Character Performance & Acting1.5%63.00C14Delivered File & Playback Integrity0.5%100.00C15Quality + Reliability2%76.14Get the complete Gen-4.5 category evidence, comparison videos, and version analysis.
If you work at Runway and want full access to the evaluation, please reach out directly to receive a full complimentary report.This Video
100.0This Video · Assessment
This video fulfills its explicit, observable instructions with near-complete consistency.
This Category
76.060 videos were rated for this category
This Category · Assessment
Most prompt requirements are fulfilled, though complex combinations or secondary details recur as limitations.
Models that scored similarly on this category
This Video
49.0This Video · Assessment
This video maintains subjects, objects, spatial relationships, environment, and lighting over time only partly reliably, with frequent important failures.
This Category
78.160 videos were rated for this category
This Category · Assessment
Scenes remain stable overall, though bounded identity, lighting, or relationship drift recurs.
Models that scored similarly on this category
This Video
50.0This Video · Assessment
The intended physical event is recognizable, but repeated important defects leave it only partly convincing.
This Category
38.817 videos were rated for this category
This Category · Assessment
Central physical events occur, but their behavior is often clearly implausible.
Models that scored similarly on this category
This Video
90.0This Video · Assessment
Highly convincing, with one very small localized defect.
This Category
67.427 videos were rated for this category
This Category · Assessment
People remain believable in most cases, though uncanny or anatomical defects recur under inspection.
Models that scored similarly on this category
This Video
40.0This Video · Assessment
The requested 2D style is present, but its animation is consistently poorly executed.
This Category
47.512 videos were rated for this category
This Category · Assessment
Coherent 2D craft appears only in some cases; line, timing, pose, or medium-authenticity failures recur frequently.
Models that scored similarly on this category
This Video
50.0This Video · Assessment
The intended 3D animation is recognizable, but repeated technical or expressive defects leave it only partly convincing.
This Category
44.05 videos were rated for this category
This Category · Assessment
Convincing 3D craft appears only in some cases; articulation, contact, rendering, or staging failures recur frequently.
Models that scored similarly on this category
This Video
60.0This Video · Assessment
Coherent and competent but synthetic, mechanical, or visually underdeveloped.
This Category
70.07 videos were rated for this category
This Category · Assessment
Shot design is technically strong overall, though recurring choices are more conventional or less expressive.
Models that scored similarly on this category
This Video
50.0This Video · Assessment
Several successful individual choices are present, but repeated inconsistencies leave the overall art direction only partly resolved.
This Category
49.760 videos were rated for this category
This Category · Assessment
Strong art direction appears only in some cases; weak hierarchy, cliché, clutter, or stylistic indecision recurs frequently.
Models that scored similarly on this category
This Video
30.0This Video · Assessment
Severe anatomical, surface, gait, or behavioral failures.
This Category
61.47 videos were rated for this category
This Category · Assessment
Animals and creatures remain believable in most cases, though synthetic anatomy, motion, or behavior recurs.
Models that scored similarly on this category
This Video
100.0This Video · Assessment
This video preserves the requested objects’ identity, structure, materials, details, motion, and interactions with near-complete consistency.
This Category
74.48 videos were rated for this category
This Category · Assessment
Objects remain correct and stable overall, though recurring secondary details or interactions are imperfect.
This Video
90.0This Video · Assessment
Highly coherent, with one small localized ambiguity.
This Category
72.711 videos were rated for this category
This Category · Assessment
Events remain clear overall, though recurring logical irregularities are visible but bounded.
Models that scored similarly on this category
This Video
61.1This Video · Assessment
This video preserves requested text content, typography, placement, attachment, legibility, and temporal stability in a generally coherent way, though visible failures recur.
This Category
47.27 videos were rated for this category
This Category · Assessment
Only some text is rendered reliably; spelling, legibility, placement, or temporal failures are frequent.
Models that scored similarly on this category
This Video
80.0This Video · Assessment
Professional, clear, and convincing, but more conventional or less nuanced.
This Category
63.021 videos were rated for this category
This Category · Assessment
Acting is generally believable, but synthetic timing, expression, attention, or specificity recurs.
Models that scored similarly on this category
megaton.aimegaton.aimegaton.aimegaton.aimegaton.aimegaton.aimegaton.aimegaton.aimegaton.aimegaton.aimegaton.aimegaton.aimegaton.aimegaton.aiMeasuring video model quality and performance
Scores are from the active Megaton v-benchmark v2.0 publication. See how our scores are calculated in our Methodology.
The Megaton Index is the published 0–100 weighted summary for v-benchmark v2. This public report shows the model's Index and category scores; the full calibration evidence, decision records, and scoring implementation are reserved for the full report.
Read the full methodology →
