Data points: 117
Model compare
The readout for DeepSeek V3.2 and DeepSeek V4 Pro, before the detailed comparison sheet.
Weighted outcome: DeepSeek V4 Pro. Benchmark capability categories carry 80%, while price, API performance, and availability carry 20%.
Decision read
DeepSeek V4 Pro
DeepSeek V4 Pro has the higher weighted result; Model A / B score 38.7 to 61.3.
Evidence depth
117 data points
Includes 23 benchmark rows, 1 audit samples, and 7 provider examples.
Selection signal
Start with DeepSeek V4 Pro
The charts below split 31 high-signal samples across speed, scores, and audit health.
Switch either side of this report to compare another model with the same LMSpeed data pipeline.
Select a different model to open a new comparison URL.
This report only uses LMSpeed data for DeepSeek V3.2 and DeepSeek V4 Pro: pricing, speed aggregates, third-party benchmark scores, and shared provider samples.
| Model compare | DeepSeek V3.2 | DeepSeek V4 Pro |
|---|---|---|
| Overall leader | Contender | Leading |
| Weighted overall score | 38.7 pts | 61.3 pts |
| Benchmark category leads | 2 categories | 4 categories |
| Operational advantages | Cheapest input price, Average speed, First-token latency | Free providers, Provider coverage |
| Context window | 163.8K tokens | 1.0M tokens |
| Max output | 65.5K tokens | 384K tokens |
| Modalities | Input Text Output Text | Input Text |
The overall result weights benchmark capability categories at 80% and price, API speed/latency, and availability at 20%. Recent test volume does not affect the winner, and missing benchmark categories are excluded.
| Model compare | DeepSeek V3.2 | DeepSeek V4 Pro |
|---|---|---|
| Developer | DeepSeek | DeepSeek |
| Released | Dec 2025 | Apr 2026 |
| Parameters | No data | No data |
| Tokenizer | DeepSeek | DeepSeek |
| Knowledge cutoff | No data | No data |
| OpenRouter ID | deepseek/deepseek-v3.2 | deepseek/deepseek-v4-pro |
| References | No data | No data |
This report only uses LMSpeed data for DeepSeek V3.2 and DeepSeek V4 Pro: pricing, speed aggregates, third-party benchmark scores, and shared provider samples.
DeepSeek V3.2
DeepSeek V3.2 is stronger in benchmark categories (Coding, Math) and operational dimensions (Cheapest input price, Average speed, First-token latency).
DeepSeek V4 Pro
DeepSeek V4 Pro is stronger in benchmark categories (Agents, Reasoning, Knowledge, Instruction following) and operational dimensions (Free providers, Provider coverage).
Third-party benchmark profile synced into LMSpeed; only metrics available for both models are shown.
Compare benchmark category scores on a 0-100 scale. Select a category to inspect the gap.
Avg. score
DeepSeek V3.2
46.4
Avg. score
DeepSeek V4 Pro
48.2
DeepSeek V4 Pro leads by 13.1
DeepSeek V3.2 leads by 2.2
DeepSeek V4 Pro leads by 2.4
DeepSeek V4 Pro leads by 3.4
DeepSeek V3.2 leads by 13.1
No data
No data
DeepSeek V4 Pro leads by 7.1
Metric-level scores with benchmark source, rank depth, confidence, error, and evaluation date where available.
DeepSeek V3.2
winner$0.420/M
Rank #18/163 · confidence 4
DeepSeek V4 Pro
$0.870/M
Rank #35/163 · confidence 4
DeepSeek V3.2
winner$0.315/M
Rank #27/163 · confidence 4
DeepSeek V4 Pro
$0.544/M
Rank #48/163 · confidence 4
DeepSeek V3.2
winner$0.280/M
Rank #44/163 · confidence 4
DeepSeek V4 Pro
$0.435/M
Rank #67/163 · confidence 4
DeepSeek V3.2
86.2%
Rank #15/125 · confidence 4
DeepSeek V4 Pro
winner87.5%
Rank #9/125 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
0.9
Rank #60/73 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner12.9
Rank #17/73 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
84.0%
Rank #52/187 · confidence 4
DeepSeek V4 Pro
winner90.5%
Rank #17/187 · confidence 4
DeepSeek V3.2
22.2%
Rank #49/186 · confidence 4
DeepSeek V4 Pro
winner33.5%
Rank #23/186 · confidence 4
DeepSeek V3.2
39.0
Rank #76/89 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner66.3
Rank #38/89 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
winner68.6
Rank #7/64 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
56.3
Rank #37/64 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
5.1
Rank #29/33 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner49.9
Rank #8/33 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
38.7
Rank #65/96 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner50.0
Rank #22/96 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
38.9%
Rank #81/184 · confidence 4
DeepSeek V4 Pro
winner46.4%
Rank #30/184 · confidence 4
DeepSeek V3.2
40.4
Rank #36/56 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner82.0
Rank #4/56 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
93.5
Rank #5/90 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner94.0
Rank #4/90 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
24.7
Rank #67/99 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner44.3
Rank #16/99 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
24.2
Rank #62/90 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner43.3
Rank #18/90 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
10.5
Rank #69/96 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner35.9
Rank #19/96 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
75.1
Rank #70/96 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner88.8
Rank #29/96 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
1206.0
Rank #42/65 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner1267.0
Rank #26/65 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
49.0
Rank #58/86 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner76.5
Rank #12/86 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
78.9
Rank #56/84 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner96.2
Rank #10/84 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
40.2
Rank #28/28 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner59.8
Rank #14/28 · confidence 2 · eval date 2026-04-24
DeepSeek V3.2
29.6
Rank #48/50 · confidence 1 · eval date 2025-12-01
DeepSeek V4 Pro
winner50.3
Rank #23/50 · confidence 2 · eval date 2026-04-24
Latest completed audits from shared providers, with four safety and integrity score groups plus report links.
| Provider | DeepSeek V3.2 | DeepSeek V4 Pro |
|---|---|---|
Winner: DeepSeek V4 Pro | DeepSeek V3.2 deepseek-v3.2 No audit yet | DeepSeek V4 Pro deepseek-v4-pro Audit score 87 788486100 |
Speed aggregates and input/output pricing share each provider row for real API selection and migration cost checks.
| Provider | DeepSeek V3.2 | DeepSeek V4 Pro |
|---|---|---|
126 tests | DeepSeek V3.2 speed / latency 26 tok/s / 11125ms input / output No data | DeepSeek V4 Pro speed / latency N/A / N/A input / output No data |
80 tests | DeepSeek V3.2 speed / latency 29 tok/s / 3036ms input / output No data | DeepSeek V4 Pro speed / latency N/A / N/A input / output No data |
65 tests | DeepSeek V3.2 speed / latency 27 tok/s / 1295ms input / output No data | DeepSeek V4 Pro speed / latency 35 tok/s / 7489ms input / output No data |
55 tests | DeepSeek V3.2 speed / latency 235 tok/s / 1787ms input / output No data | DeepSeek V4 Pro speed / latency 27 tok/s / 51460ms input / output No data |
55 tests | DeepSeek V3.2 speed / latency 18 tok/s / 4488ms input / output No data | DeepSeek V4 Pro speed / latency 21 tok/s / 1345ms input / output No data |
DeepSeek V3.2 deepseek-v3.2 speed / latency No data input / output $0/M/$0/M | DeepSeek V4 Pro deepseek-v4-pro speed / latency No data input / output $0.118/M | |
DeepSeek V3.2 deepseek-v3.2 speed / latency No data input / output $0/request | DeepSeek V4 Pro deepseek-v4-pro speed / latency No data input / output $0/request |
Weighted outcome: DeepSeek V4 Pro. Benchmark capability categories carry 80%, while price, API performance, and availability carry 20%.
Output
| Features | Text inputText outputTool callingStructured outputsJSON modeReasoning | Text inputText outputTool callingStructured outputsJSON modeReasoning |
|---|
