Model comparison
Compare models or reasoning efforts by score, evaluation time and estimated cost.
Available modelsSelected 2 / 3
Measured results
Differences are relative to the first column
deepseek-v4-proAesthetics +23 pts · Reasoning −10 pts
| Metric | step-5-preview · default | deepseek-v4-pro · max |
|---|---|---|
| Overall | 65.20 | 65.15−0.05 pts |
| Backend engineering | 61 | 67+6 pts |
| Frontend interaction | 71 | 63−8 pts |
| Knowledge & reasoning | 75 | 65−10 pts |
| Visual aesthetics | 47 | 70+23 pts |
| Maintainability | 55 | 59+4 pts |
Cost & time
Complete evaluation| Metric | step-5-preview · default | deepseek-v4-pro · max |
|---|---|---|
| Reference cost | $1.24 | $1.15−$0.084 |
| Evaluation time | 98m 22s | 93m 15s−5.2% |
Dimension breakdown
Expand to inspect each subaxis.
Backend engineeringSubscores
| Scoring dimension | step-5-previewdefault | deepseek-v4-promax |
|---|---|---|
| Contracts and boundaries | 9 / 20 | 9 / 20 |
| Retries and failures | 16 / 20 | 17 / 20 |
| CI and adversarial review | 8 / 20 | 12 / 20 |
| Transactions and consistency | 16 / 20 | 13 / 20 |
| Caching and regression | 12 / 20 | 16 / 20 |
Frontend interactionSubscores
| Scoring dimension | step-5-previewdefault | deepseek-v4-promax |
|---|---|---|
| Layout and adaptation | 18 / 20 | 17 / 20 |
| Interaction behavior | 18 / 20 | 16 / 20 |
| State continuity | 12 / 20 | 8 / 20 |
| Accessibility and feedback | 10 / 15 | 7 / 15 |
| State validation and regression | 13 / 25 | 15 / 25 |
Knowledge & reasoningSubscores
| Scoring dimension | step-5-previewdefault | deepseek-v4-promax |
|---|---|---|
| Mathematical reasoning | 10 / 15 | 15 / 15 |
| Computer science & AI | 10 / 15 | 10 / 15 |
| Science and engineering | 10 / 10 | 5 / 10 |
| Life sciences and medicine | 20 / 20 | 15 / 20 |
| Humanities and general knowledge | 25 / 40 | 20 / 40 |
Visual aestheticsSubscores
| Scoring dimension | step-5-previewdefault | deepseek-v4-promax |
|---|---|---|
| Composition | 15 / 20 | 19 / 20 |
| Spacing and alignment | 14 / 20 | 16 / 20 |
| Typography and expression | 15 / 20 | 18 / 20 |
| Color and surfaces | 18 / 20 | 19 / 20 |
| Control styling | 16 / 20 | 17 / 20 |
Subscores retain their original scale; the axis score uses a consistent conversion.
MaintainabilitySubscores
| Scoring dimension | step-5-previewdefault | deepseek-v4-promax |
|---|---|---|
| Semantic clarity | 15 / 20 | 15 / 20 |
| Control-flow organization | 19 / 20 | 20 / 20 |
| Locality of changes | 15 / 20 | 15 / 20 |
| State and lifecycle | 16 / 20 | 17 / 20 |
| Implementation economy | 17 / 20 | 17 / 20 |
Subscores retain their original scale; the axis score uses a consistent conversion.