DeepSeek V3 vs DeepSeek V3.2
Updated October 21, 2026. 9 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
DeepSeek V3 scores 70.2 and DeepSeek V3.2 54.4 on the Vibe Coding Index, but no coding benchmark measured both, so the gap is directional, not a ranking.
Pick DeepSeek V3.2 when output tokens cost 2.4x less, or when what matters is Design Arena: websites or Design Arena: UI components, where it leads.
- Shared results
- 9
- DeepSeek V3 only
- 4
- DeepSeek V3.2 only
- 66
- Like-for-like categories
- 2 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, four public coding benchmarks
No call
70.2 against 54.4, from different benchmarks with none in common.
Directional only
Intelligence
AA Intelligence Index
No call
Only DeepSeek V3.2 has a AA Intelligence Index result, 21.5.
Directional only
Reasoning
GPQA Diamond
No call
Only DeepSeek V3.2 has a GPQA Diamond result, 80.3%.
Directional only
Math
AIME
No call
Only DeepSeek V3.2 has a AIME result, 84.6%.
Directional only
Knowledge
MMLU Pro
No call
Only DeepSeek V3.2 has a MMLU Pro result, 84.9%.
Directional only
Web design
Design Arena: websites
DeepSeek V3.2
DeepSeek V3.2 leads, 1175 against 1120.
Same board, both measured
Long documents
Listed context window
No call
Both take 164K tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
DeepSeek V3.2
DeepSeek V3 $0.0008, DeepSeek V3.2 $0.0005 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only DeepSeek V3.2 has the number, $0.99.
Directional only
At a glance
- Coding index
- DeepSeek V3 70.2 from 1 benchmark, DeepSeek V3.2 54.4 from 2 benchmarks.
- Widest category
- General, 18.3 points apart on average over 2 shared benchmarks.
- Widest single gap
- DTBench: DeepSeek V3 62.7% against DeepSeek V3.2 85.6%, 22.9 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V3 $1.0287 against DeepSeek V3.2 $0.42 per million output tokens, 2.4x apart.
- Context window
- Both take 164K tokens.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | DeepSeek V3 | Bars | DeepSeek V3.2 | Ties |
|---|---|---|---|---|
| Web design | 0 | 7 | ||
| General | 0 | 2 | ||
| All shared boards | 0 | 9 |
CodingDirectional onlyFour of these boards feed the Vibe Coding Index16 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Leader |
|---|---|---|---|
| Aider polyglot | 70.2% | Not measured | Only DeepSeek V3 measured |
| SWE-bench Verified | Not measured | 70.0% | Only DeepSeek V3.2 measured |
| SWE-bench Lite | 30.7% | Not measured | Only DeepSeek V3 measured |
| SWE-bench Multilingual | Not measured | 59.0% | Only DeepSeek V3.2 measured |
| LMArena WebDev | Not measured | 1362 | Only DeepSeek V3.2 measured |
| AA Coding Index | Not measured | 44.2 | Only DeepSeek V3.2 measured |
| LiveCodeBench (Vals AI) | Not measured | 80.7% | Only DeepSeek V3.2 measured |
| SWE-bench Verified (Vals AI) | Not measured | 67.6% | Only DeepSeek V3.2 measured |
| Vibe Code Bench (Vals AI) | Not measured | 5.1% | Only DeepSeek V3.2 measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 86.2% | Only DeepSeek V3.2 measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 35.6% | Only DeepSeek V3.2 measured |
| Terminal-Bench | Not measured | 39.6% | Only DeepSeek V3.2 measured |
| LMArena Text: coding | Not measured | 1453 | Only DeepSeek V3.2 measured |
| LMArena Text: software and IT | Not measured | 1449 | Only DeepSeek V3.2 measured |
| LMArena WebDev: React | Not measured | 1350 | Only DeepSeek V3.2 measured |
| LMArena WebDev: HTML | Not measured | 1371 | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 22 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Leader |
|---|---|---|---|
| AA Intelligence Index | Not measured | 21.5 | Only DeepSeek V3.2 measured |
| Epoch Capabilities Index | Not measured | 146.1 | Only DeepSeek V3.2 measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index3 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Leader |
|---|---|---|---|
| τ²-Bench Telecom | Not measured | 90.6% | Only DeepSeek V3.2 measured |
| APEX-Agents | Not measured | 21.3% | Only DeepSeek V3.2 measured |
| Vending-Bench 2 | Not measured | $1,034 | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 36 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index9 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 80.3% | Only DeepSeek V3.2 measured |
| GPQA Diamond (Artificial Analysis run) | Not measured | 84.0% | Only DeepSeek V3.2 measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 24.6% | Only DeepSeek V3.2 measured |
| GPQA Diamond (Epoch AI run) | 71.2% | Not measured | Only DeepSeek V3 measured |
| ARC-AGI | Not measured | 57.0% | Only DeepSeek V3.2 measured |
| ARC-AGI-2 | Not measured | 4.0% | Only DeepSeek V3.2 measured |
| LMArena Text: hard prompts | Not measured | 1434 | Only DeepSeek V3.2 measured |
| LMArena Text: expert | Not measured | 1436 | Only DeepSeek V3.2 measured |
| LMArena Text: hard prompts in English | Not measured | 1446 | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 12 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index7 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Leader |
|---|---|---|---|
| AIME | Not measured | 84.6% | Only DeepSeek V3.2 measured |
| MGSM (Vals AI) | Not measured | 86.0% | Only DeepSeek V3.2 measured |
| AA Math Index | Not measured | 92.0 | Only DeepSeek V3.2 measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 92.0% | Only DeepSeek V3.2 measured |
| OTIS Mock AIME | 48.9% | Not measured | Only DeepSeek V3 measured |
| LMArena Text: math | Not measured | 1435 | Only DeepSeek V3.2 measured |
| LMArena Text: mathematical work | Not measured | 1440 | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 6 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index11 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 84.9% | Only DeepSeek V3.2 measured |
| LegalBench (Vals AI) | Not measured | 76.1% | Only DeepSeek V3.2 measured |
| CaseLaw (Vals AI) | Not measured | 55.4% | Only DeepSeek V3.2 measured |
| CorpFin (Vals AI) | Not measured | 51.0% | Only DeepSeek V3.2 measured |
| TaxEval (Vals AI) | Not measured | 68.2% | Only DeepSeek V3.2 measured |
| MedQA (Vals AI) | Not measured | 93.9% | Only DeepSeek V3.2 measured |
| MMLU Pro (Artificial Analysis run) | Not measured | 86.2% | Only DeepSeek V3.2 measured |
| LMArena Text: science | Not measured | 1446 | Only DeepSeek V3.2 measured |
| LMArena Text: medicine and healthcare | Not measured | 1440 | Only DeepSeek V3.2 measured |
| LMArena Text: legal and government | Not measured | 1425 | Only DeepSeek V3.2 measured |
| LMArena Text: business and finance | Not measured | 1410 | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 6 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index07
| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Leader |
|---|---|---|---|
| Design Arena: websites | 1120 | 1175 | DeepSeek V3.2 |
| Design Arena: UI components | 1093 | 1149 | DeepSeek V3.2 |
| Design Arena: game dev | 1054 | 1132 | DeepSeek V3.2 |
| Design Arena: data viz | 1089 | 1158 | DeepSeek V3.2 |
| Design Arena: 3D | 1087 | 1133 | DeepSeek V3.2 |
| Design Arena: all code categories | 1108 | 1164 | DeepSeek V3.2 |
| Design Arena: SVG | 976 | 1032 | DeepSeek V3.2 |
| Design Arena: ASCII art | Not measured | 1082 | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Leader |
|---|---|---|---|
| IFBench | Not measured | 60.7% | Only DeepSeek V3.2 measured |
| Long context reasoning (AA-LCR) | Not measured | 73.3% | Only DeepSeek V3.2 measured |
| DTBench | 62.7% | 85.6% | DeepSeek V3.2 |
| LMCA | 15.2% | 28.8% | DeepSeek V3.2 |
| CL-bench | Not measured | 12.4% | Only DeepSeek V3.2 measured |
| CL-bench Life | Not measured | 7.4% | Only DeepSeek V3.2 measured |
| LMArena Text | Not measured | 1425 | Only DeepSeek V3.2 measured |
| LMArena Text: instruction following | Not measured | 1413 | Only DeepSeek V3.2 measured |
| LMArena Text: creative writing | Not measured | 1400 | Only DeepSeek V3.2 measured |
| LMArena Text: English | Not measured | 1439 | Only DeepSeek V3.2 measured |
| LMArena Text: non-English | Not measured | 1409 | Only DeepSeek V3.2 measured |
| LMArena Text: Chinese | Not measured | 1455 | Only DeepSeek V3.2 measured |
| LMArena Text: German | Not measured | 1413 | Only DeepSeek V3.2 measured |
| LMArena Text: Japanese | Not measured | 1374 | Only DeepSeek V3.2 measured |
| LMArena Text: Korean | Not measured | 1370 | Only DeepSeek V3.2 measured |
| LMArena Text: French | Not measured | 1429 | Only DeepSeek V3.2 measured |
| LMArena Text: Spanish | Not measured | 1424 | Only DeepSeek V3.2 measured |
| LMArena Text: Russian | Not measured | 1424 | Only DeepSeek V3.2 measured |
| LMArena Text: Polish | Not measured | 1419 | Only DeepSeek V3.2 measured |
| LMArena Text: multi-turn | Not measured | 1427 | Only DeepSeek V3.2 measured |
| LMArena Text: longer queries | Not measured | 1428 | Only DeepSeek V3.2 measured |
| LMArena Text: writing and language | Not measured | 1407 | Only DeepSeek V3.2 measured |
| LMArena Text: entertainment and media | Not measured | 1395 | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 23 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
DeepSeek V3.2 costs less
Repository review
50K input + 3K output tokens
DeepSeek V3.2 costs less
Agent loop
200K input + 10K output tokens
DeepSeek V3.2 costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | DeepSeek V3 | DeepSeek V3.2 |
|---|---|---|
| Maker | DeepSeek | DeepSeek |
| Weights | Open weights | Open weights |
| Released | December 26, 2024 | December 1, 2025 |
| Context window | 164K | 164K |
| Input $/Mtok | $0.2574 | $0.28 |
| Output $/Mtok | $1.0287 | $0.42 |
| Coding index | 70.2 | 54.4 |
| Independent measurements behind it | 1 | 2 |
| Cost per solved task | Not computable | $0.99 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V3 better than DeepSeek V3.2 for coding?
DeepSeek V3 scores 70.2 and DeepSeek V3.2 54.4 on the Vibe Coding Index, but no coding benchmark measured both, so the gap is directional, not a ranking. Pick DeepSeek V3.2 when output tokens cost 2.4x less, or when what matters is Design Arena: websites or Design Arena: UI components, where it leads.
Which is cheaper, DeepSeek V3 or DeepSeek V3.2?
DeepSeek V3 costs $0.2574 per million input tokens and $1.0287 per million output tokens; DeepSeek V3.2 costs $0.28 and $0.42. Cost per solved task is known for DeepSeek V3.2 only.
Which has the bigger context window, DeepSeek V3 or DeepSeek V3.2?
Neither: both take 164K tokens.
How many benchmarks is the DeepSeek V3 vs DeepSeek V3.2 comparison based on?
9 benchmarks have a result for both models as of October 21, 2026. DeepSeek V3's coding index rests on 1 independent benchmark and DeepSeek V3.2's on 2. Where only one model has a result, the page says so instead of naming a winner.