DeepSeek-V4-FlashvsDeepSeek V3.2
Across 9 shared benchmarks, DeepSeek-V4-Flash leads overall: DeepSeek-V4-Flash wins 6, DeepSeek V3.2 wins 3, with 0 ties and an average score difference of +73.16.
DeepSeek-V4-Flash
DeepSeek-AI · 2026-04-24 · Reasoning model
DeepSeek V3.2
DeepSeek-AI · 2025-12-01 · Reasoning model
DeepSeek-V4-Flash6 wins(67%)(33%)3 winsDeepSeek V3.2
Benchmark scores
Grouped by capability, sorted by largest gap within each. 9 shared benchmarks.
Coding and Software Engineer
DeepSeek-V4-Flash 3/4| Benchmark | DeepSeek-V4-Flash | DeepSeek V3.2 | Diff |
|---|---|---|---|
| CodeForces | 3,0523 / 20最高（无工具） | 2,38611 / 20Thinking (No Tools) | +666 |
| LiveCodeBench | 55.2086 / 126Normal (No Tools) | 83.3023 / 126Thinking (No Tools) | -28.10 |
| SWE-Bench Pro - Public | 49.1047 / 57Normal (With Tools) | 40.9052 / 57Thinking (No Tools) | +8.20 |
| SWE-bench Verified | 73.7045 / 114Normal (With Tools) | 73.1050 / 114 | +0.60 |
General Knowledge
Even 2/2| Benchmark | DeepSeek-V4-Flash | DeepSeek V3.2 | Diff |
|---|---|---|---|
| HLE | 8.10164 / 181Normal (No Tools) | 25.10110 / 181Thinking (No Tools) | -17 |
| LiveBench | 67.2549 / 115Normal (No Tools) | 51.8487 / 115Normal (No Tools) | +15.41 |
AI Agent - Information Search
DeepSeek-V4-Flash 1/1| Benchmark | DeepSeek-V4-Flash | DeepSeek V3.2 | Diff |
|---|---|---|---|
| BrowseComp | 73.2028 / 54极高强度思考（工具） | 51.4042 / 54Thinking (No Tools) | +21.80 |
AI Agent - Tool Usage
DeepSeek-V4-Flash 1/1| Benchmark | DeepSeek-V4-Flash | DeepSeek V3.2 | Diff |
|---|---|---|---|
| Terminal Bench 2.0 | 49.1036 / 48Normal (With Tools) | 46.4041 / 48 | +2.70 |
General Evaluation
DeepSeek V3.2 1/1| Benchmark | DeepSeek-V4-Flash | DeepSeek V3.2 | Diff |
|---|---|---|---|
| GPQA Diamond | 71.20151 / 226Normal (No Tools) | 82.40106 / 226Thinking (No Tools) | -11.20 |
Specs
| Field | DeepSeek-V4-Flash | DeepSeek V3.2 |
|---|---|---|
| Publisher | DeepSeek-AI | DeepSeek-AI |
| Release date | 2026-04-24 | 2025-12-01 |
| Model type | Reasoning model | Reasoning model |
| Architecture | MoE | MoE |
| Parameters | 284B | 671B |
| Context length | 1M | 128K |
| Max output | 384K | 8K |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | DeepSeek-V4-Flash | DeepSeek V3.2 |
|---|---|---|
| Text input | $0.14 / 1M tokens | $0.28 / 1M tokens |
| Text output | $0.28 / 1M tokens | $0.42 / 1M tokens |
| Cache read | $0.0028 / 1M tokens | $0.028 / 1M tokens |
| Cache write | Not public | $0.28 / 1M tokens |
Summary
- DeepSeek-V4-Flashleads in:Coding and Software Engineer (3/4), AI Agent - Information Search (1/1), AI Agent - Tool Usage (1/1)
- DeepSeek V3.2leads in:General Evaluation (1/1)
- Tied in:General Knowledge
On average across the 9 shared benchmarks, DeepSeek-V4-Flash scores 73.16 higher.
Largest single-benchmark gap: CodeForces — DeepSeek-V4-Flash 3,052 vs DeepSeek V3.2 2,386 (+666).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.