Model comparison
DeepSeek V3.2 vs DeepSeek V4 Flash
Head-to-head evidence from 3 shared benchmark results across 2 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.
Public leaderboard positions: DeepSeek V3.2 #82 (Supported); DeepSeek V4 Flash #61 (Estimated). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.
Evidence parity. DeepSeek V3.2 and DeepSeek V4 Flash share 3 comparable benchmark results. 2 of 8 categories are comparable. 16 results are unique to DeepSeek V3.2; 19 to DeepSeek V4 Flash.
Updated July 23, 2026- Shared results
- 3
- DeepSeek V3.2 only
- 16
- DeepSeek V4 Flash only
- 19
- Comparable categories
- 2 / 8
Pick DeepSeek V4 Flash if you want the stronger benchmark profile. DeepSeek V3.2 only becomes the better choice if its workflow or ecosystem matters more than the raw scoreboard.
Confidence note. This is a partial-evidence comparison with 3 shared benchmark results across 2 evidence categories; 2 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.
Why this result
DeepSeek V4 Flash is clearly ahead on the BenchAlign aggregate, 58.88 to 55.4. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
DeepSeek V4 Flash's sharpest advantage is in mathematics, where it averages 40.8 against 17.1.
DeepSeek V3.2 is also the more expensive model on tokens at $0.28 input / $0.42 output per 1M tokens, versus $0.14 input / $0.28 output per 1M tokens for DeepSeek V4 Flash. DeepSeek V4 Flash gives you the larger context window at 1M, compared with 128K for DeepSeek V3.2.
Category breakdown
Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.
| Category | DeepSeek V3.2 | Δ | DeepSeek V4 Flash |
|---|---|---|---|
| Math | DeepSeek V3.217.1 | Margin→ 23.7 | DeepSeek V4 Flash40.8 |
| Coding | DeepSeek V3.260.9 | Margin→ 3.3 | DeepSeek V4 Flash64.2 |
| Agentic | DeepSeek V3.2Not measured | MarginNo overlap | DeepSeek V4 Flash49.1 |
| Knowledge | DeepSeek V3.2Not measured | MarginNo overlap | DeepSeek V4 Flash38.8 |
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | DeepSeek V3.2 | DeepSeek V4 Flash | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | DeepSeek V3.2$0.28 input / $0.42 output | DeepSeek V4 Flash$0.14 input / $0.28 output | DeepSeek V4 Flash has the lower combined listed price. |
| Generation speedtokens per second | DeepSeek V3.235 tok/s | DeepSeek V4 FlashNot available | A complete speed comparison is not available. |
| First-answer latencyseconds to first token | DeepSeek V3.23.75 s | DeepSeek V4 FlashNot available | A complete latency comparison is not available. |
| Context windowmaximum listed tokens | DeepSeek V3.2128K | DeepSeek V4 Flash1M | DeepSeek V4 Flash lists the larger context window. |
Benchmark Deep Dive
Agentic7 benchmarks
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Flash | Result |
|---|---|---|---|
| Claw-EvalSource | 40.2% | 57.8% | DeepSeek V4 Flash leads |
| VITA-BenchSource | 18.5% | — | Not comparable |
| τ²-bench resultsSource | 78.9% | — | Not comparable |
| Gert LabsSource | 29.57% | 54.35% | DeepSeek V4 Flash leads |
| Terminal-Bench 2.0Source | — | 49.1% | Not comparable |
| MCP AtlasSource | — | 64% | Not comparable |
| ToolathlonSource | — | 40.7% | Not comparable |
CodingDeepSeek V4 Flash wins7 benchmarks
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Flash | Result |
|---|---|---|---|
| SWE-RebenchSource | 60.9% | — | Not comparable |
| React Native EvalsSource | 71.5% | — | Not comparable |
| AA-SciCodeSource | 38.7% | — | Not comparable |
| SWE-bench VerifiedSource | — | 73.7% | Not comparable |
| SWE-bench ProSource | — | 49.1% | Not comparable |
| SWE MultilingualSource | — | 69.7% | Not comparable |
| Terminal-Bench 2.0Source | — | 49.1% | Not comparable |
Reasoning4 benchmarks
Knowledge12 benchmarks
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Flash | Result |
|---|---|---|---|
| Artificial Analysis Intelligence IndexSource | 24.7% | — | Not comparable |
| AA-GPQA DiamondSource | 75.1% | — | Not comparable |
| AA-HLESource | 10.5% | — | Not comparable |
| AA-Omniscience IndexSource | -46.7% | — | Not comparable |
| AA-Omniscience AccuracySource | 24.2% | — | Not comparable |
| AA-Omniscience Hallucination RateSource | 93.5% | — | Not comparable |
| MMLU-ProSource | — | 83% | Not comparable |
| SimpleQASource | — | 23.1% | Not comparable |
| Chinese-SimpleQASource | — | 71.5% | Not comparable |
| GPQASource | — | 71.2% | Not comparable |
| GPQA-DSource | — | 71.2% | Not comparable |
| HLESource | — | 8.1% | Not comparable |
MathDeepSeek V4 Flash wins6 benchmarks
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Flash | Result |
|---|---|---|---|
| FrontierMath v2 (Tiers 1-3)Source | 22.100% | — | Not comparable |
| FrontierMath v2 (Tier 4)Source | 2.100% | — | Not comparable |
| HMMT Feb 2026Source | — | 40.8% | Not comparable |
| IMOAnswerBenchSource | — | 41.9% | Not comparable |
| ApexSource | — | 1.0% | Not comparable |
| Apex ShortlistSource | — | 9.3% | Not comparable |
Multimodal1 benchmarks
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Flash | Result |
|---|---|---|---|
| Design Arena WebsiteSource | 1204 | 1238 | DeepSeek V4 Flash leads |
Inst. Following1 benchmarks
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Flash | Result |
|---|---|---|---|
| AA-IFBenchSource | 49.0% | — | Not comparable |
Frequently Asked Questions (3)
Which is better, DeepSeek V3.2 or DeepSeek V4 Flash?
DeepSeek V4 Flash is ahead on BenchLM's BenchAlign leaderboard, 58.88 to 55.4.
Which is better for coding, DeepSeek V3.2 or DeepSeek V4 Flash?
DeepSeek V4 Flash has the edge for coding in this comparison, averaging 64.2 versus 60.9. DeepSeek V3.2 stays close enough that the answer can still flip depending on your workload.
Which is better for math, DeepSeek V3.2 or DeepSeek V4 Flash?
DeepSeek V4 Flash has the edge for math in this comparison, averaging 40.8 versus 17.1. DeepSeek V3.2 stays close enough that the answer can still flip depending on your workload.
Related Comparisons
Explore More
Choose a model with this week’s evidence
Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.
One email each week. Unsubscribe anytime.