DeepSeek V4
DeepSeek V4 has 4 tested configurations. Its highest published overall score is 65.15 at max. Inspect dimension scores, evaluation time, and reference cost below.
Overall score65.15
deepseek-v4-pro · max
- Backend engineering
- 67
- Frontend interaction
- 63
- Knowledge & Reasoning
- 65
- Visual aesthetics
- 70
- Maintainability
- 59
- Evaluation time
- 93m 15s
- Reference cost
- $1.15
Complete evaluation · API reference cost
Results by reasoning effort
Swipe to see all results →
| Model / effort | Overall | Backend engineering | Frontend interaction | Knowledge & Reasoning | Visual aesthetics | Maintainability | Total eval. time | Total ref. cost | Compare |
|---|---|---|---|---|---|---|---|---|---|
| deepseek-v4-pro | 63.60 | 61 | 54 | 75 | 59 | 55 | 61m 4s | $0.98Partial cost | |
| deepseek-v4-flash | 60.50 | 58 | 76 | 55 | 61 | 53 | 98m 25s | $0.42 | |
| deepseek-v4-pro | 65.15 | 67 | 63 | 65 | 70 | 59 | 93m 15s | $1.15 | |
| deepseek-v4-flash | 56.10 | 54 | 61 | 55 | 61 | 47 | 89m 12s | $0.60 |
Score breakdown
- Overall score
- 65.15
- Evaluation time
- 93m 15s
- Reference cost
- $1.15
Service logic, error handling, and test verification.
Contracts and boundaries9 / 20
API contracts, input validation, and boundary behavior.
Retries and failures17 / 20
Error recovery, retry conditions, and failure handling.
CI and adversarial review12 / 20
Verification plans, risk detection, and audit evidence.
Transactions and consistency13 / 20
State transitions, concurrency control, and data consistency.
Caching and regression16 / 20
Cache invalidation, data updates, and regression coverage.
What changes with higher reasoning effort
Compare overall scores, cost, and time across neighboring tested efforts.
high max
- Overall
- +1.6 pts 63.60 → 65.15
- Reference cost
- — — → $1.15
- Evaluation time
- +52.7% 61m 4s → 93m 15s
Data snapshot published · 10/27/2026Scoring & cost notes