Model comparison
Selected 2 / 5See the differences. Choose what matters to you.
Available modelsSelected 2 / 5
Differences vs. DeepSeek V4 Pro · max
Score, cost & time differences
Current baseline · DeepSeek V4 Pro · max
63.45Overall
Compared with DeepSeek V4 Pro · max
- Step-5 Preview · default
Overall: −0.55 ptsCost $1.24 ($0.084 more)Evaluation time 98m 22s (5m 7s more)
Differences reflect these evaluation tasks. Cost and time cover the same dimension, not a typical conversation. Scoring & cost notes
| Metric | DeepSeek V4 Pro · max | Step-5 Preview · default |
|---|---|---|
| Overall | DeepSeek V4 Promax 63.45 | Step-5 Previewdefault 62.90−0.55 pts |
| Backend engineering | DeepSeek V4 Promax 64 | Step-5 Previewdefault 56−8 pts |
| Frontend interaction | DeepSeek V4 Promax 60 | Step-5 Previewdefault 61+1 pts |
| Knowledge & reasoning | DeepSeek V4 Promax 65 | Step-5 Previewdefault 75+10 pts |
| Visual aesthetics | DeepSeek V4 Promax 70 | Step-5 Previewdefault 47−23 pts |
| Maintainability | DeepSeek V4 Promax 59 | Step-5 Previewdefault 55−4 pts |
Complete evaluation · Evaluation consumption, not a monthly subscription fee
Cost & time
Complete evaluation| Metric | DeepSeek V4 Pro · max | Step-5 Preview · default |
|---|---|---|
| Cost | DeepSeek V4 Promax $1.15 | Step-5 Previewdefault $1.24+$0.084 |
| Evaluation time | DeepSeek V4 Promax 93m 15s | Step-5 Previewdefault 98m 22s+5.5% |
Capability profiles
Show up to five models. Select names to focus the chart.
Behind each score
Expand to inspect each subaxis.
Backend engineeringSubscores
| Scoring dimension | DeepSeek V4 Promax | Step-5 Previewdefault |
|---|---|---|
| Contracts and boundaries | 9 / 20 | 9 / 20 |
| Retries and failures | 17 / 20 | 16 / 20 |
| CI and adversarial review | 12 / 20 | 8 / 20 |
| Transactions and consistency | 13 / 20 | 16 / 20 |
| Caching and regression | 13 / 20 | 7 / 20 |
Frontend interactionSubscores
| Scoring dimension | DeepSeek V4 Promax | Step-5 Previewdefault |
|---|---|---|
| Layout and adaptation | 17 / 20 | 18 / 20 |
| Interaction behavior | 15 / 20 | 13 / 20 |
| State continuity | 6 / 20 | 9 / 20 |
| Accessibility and feedback | 7 / 15 | 8 / 15 |
| State validation and regression | 15 / 25 | 13 / 25 |
Knowledge & reasoningSubscores
| Scoring dimension | DeepSeek V4 Promax | Step-5 Previewdefault |
|---|---|---|
| Mathematical reasoning | 15 / 15 | 10 / 15 |
| Computer science & AI | 10 / 15 | 10 / 15 |
| Science and engineering | 5 / 10 | 10 / 10 |
| Life sciences and medicine | 15 / 20 | 20 / 20 |
| Humanities and general knowledge | 20 / 40 | 25 / 40 |
Visual aestheticsSubscores
| Scoring dimension | DeepSeek V4 Promax | Step-5 Previewdefault |
|---|---|---|
| Composition | 19 / 20 | 15 / 20 |
| Spacing and alignment | 16 / 20 | 14 / 20 |
| Typography and expression | 18 / 20 | 15 / 20 |
| Color and surfaces | 19 / 20 | 18 / 20 |
| Control styling | 17 / 20 | 16 / 20 |
Subscores retain their original scale; the axis score uses a consistent conversion.
MaintainabilitySubscores
| Scoring dimension | DeepSeek V4 Promax | Step-5 Previewdefault |
|---|---|---|
| Semantic clarity | 15 / 20 | 15 / 20 |
| Control-flow organization | 20 / 20 | 19 / 20 |
| Locality of changes | 15 / 20 | 15 / 20 |
| State and lifecycle | 17 / 20 | 16 / 20 |
| Implementation economy | 17 / 20 | 17 / 20 |
Subscores retain their original scale; the axis score uses a consistent conversion.