DeepSeek V3.2 vs DeepSeek V4 Pro 0813
DeepSeek V4 Pro 0813 leads DeepSeek V3.2 on coding, 56.1 to 30.0 on the Vibe Coding Index, and DeepSeek V3.2 costs 2.3x less per output token ($0.88 vs $1.98 per million). Updated October 22, 2026; 47 benchmarks have a result for both.
DeepSeek V3.2 or DeepSeek V4 Pro 0813: which to pick
Pick DeepSeek V4 Pro 0813 for coding: it scores 56.1 on the Vibe Coding Index against 30.0 for DeepSeek V3.2.
Pick DeepSeek V3.2 for output tokens 2.3x cheaper ($0.88 vs $1.98 per million).
- Shared results
- 47
- DeepSeek V3.2 only
- 28
- DeepSeek V4 Pro 0813 only
- 31
- Like-for-like categories
- 7 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
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, six public coding benchmarks
DeepSeek V4 Pro 0813
DeepSeek V4 Pro 0813 scores 56.1 against 30.0.
3 and 5 benchmarks behind the scores
Intelligence
AA Intelligence Index
DeepSeek V4 Pro 0813
DeepSeek V4 Pro 0813 leads, 36.0 against 21.5.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only DeepSeek V4 Pro 0813 has a AA Agentic Index result, 41.3.
Directional only
Reasoning
GPQA Diamond
DeepSeek V4 Pro 0813
DeepSeek V4 Pro 0813 leads, 92.4% against 80.3%.
Same board, both measured
Math
AIME
No call
Only DeepSeek V3.2 has a AIME result, 84.6%.
Directional only
Knowledge
MMLU Pro
DeepSeek V4 Pro 0813
DeepSeek V4 Pro 0813 leads, 87.0% against 84.9%.
Same board, both measured
Web design
Design Arena: websites
No call
Only DeepSeek V3.2 has a Design Arena: websites result, 1175.
Directional only
General
Vals Index
No call
Only DeepSeek V4 Pro 0813 has a Vals Index result, 53.0%.
Directional only
Long documents
Listed context window
DeepSeek V4 Pro 0813
DeepSeek V4 Pro 0813 takes 1.05M tokens against 164K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
DeepSeek V3.2
DeepSeek V3.2 $0.0007, DeepSeek V4 Pro 0813 $0.0016 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
DeepSeek V4 Pro 0813
DeepSeek V3.2 $0.99, DeepSeek V4 Pro 0813 $0.10.
Published attempt cost and pass rate
At a glance
- Coding index
- DeepSeek V3.2 30.0 from 3 benchmarks, DeepSeek V4 Pro 0813 56.1 from 5 benchmarks.
- Widest category
- Coding, 27.3 points apart on average over 9 shared benchmarks.
- Widest single gap
- Vibe Code Bench (Vals AI): DeepSeek V3.2 5.1% against DeepSeek V4 Pro 0813 82.3%, 77.2 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V3.2 $0.88 against DeepSeek V4 Pro 0813 $1.98 per million output tokens, 2.3x apart.
- Context window
- DeepSeek V3.2 takes 164K tokens, DeepSeek V4 Pro 0813 1.05M.
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 V3.2 | Bars | DeepSeek V4 Pro 0813 | Ties |
|---|---|---|---|---|
| Coding | 0 | 9 | ||
| Intelligence | 0 | 2 | ||
| Agentic work | 0 | 1 | ||
| Reasoning | 0 | 8 | ||
| Math | 0 | 2 | ||
| Knowledge | 0 | 8 | ||
| General | 0 | 17 | ||
| All shared boards | 0 | 47 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index09
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Pro 0813 | Leader |
|---|---|---|---|
| SWE-bench Verified | 70.0% | Not measured | Only DeepSeek V3.2 measured |
| SWE-bench Multilingual | 59.0% | Not measured | Only DeepSeek V3.2 measured |
| LMArena WebDev | 1362 | 1582 | DeepSeek V4 Pro 0813 |
| AA Coding Index | 44.2 | 68.8 | DeepSeek V4 Pro 0813 |
| LiveCodeBench (Vals AI) | 80.7% | 87.5% | DeepSeek V4 Pro 0813 |
| SWE-bench Verified (Vals AI) | 67.6% | 96.4% | DeepSeek V4 Pro 0813 |
| Terminal-Bench 4 (Vals AI) | Not measured | 14.1% | Only DeepSeek V4 Pro 0813 measured |
| Vibe Code Bench (Vals AI) | 5.1% | 82.3% | DeepSeek V4 Pro 0813 |
| Code migration (Vals AI) | Not measured | 41.5% | Only DeepSeek V4 Pro 0813 measured |
| ProgramBench (Vals AI) | Not measured | 0.0% | Only DeepSeek V4 Pro 0813 measured |
| IOI (Vals AI) | Not measured | 51.6% | Only DeepSeek V4 Pro 0813 measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 54.7% | Only DeepSeek V4 Pro 0813 measured |
| Vibe Code Bench 1-100 (Vals AI) | Not measured | 17.5% | Only DeepSeek V4 Pro 0813 measured |
| SkillsBench (Vals AI) | Not measured | 53.8% | Only DeepSeek V4 Pro 0813 measured |
| CyberBench (Vals AI) | Not measured | 72.9% | Only DeepSeek V4 Pro 0813 measured |
| LiveCodeBench (Artificial Analysis run) | 86.2% | Not measured | Only DeepSeek V3.2 measured |
| SciCode | Not measured | 51.0% | Only DeepSeek V4 Pro 0813 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 35.6% | Not measured | Only DeepSeek V3.2 measured |
| Terminal-Bench | 39.6% | Not measured | Only DeepSeek V3.2 measured |
| FrontierCode | Not measured | 28.5% | Only DeepSeek V4 Pro 0813 measured |
| WeirdML | Not measured | 66.2% | Only DeepSeek V4 Pro 0813 measured |
| ALE-Bench | Not measured | 1403.2 | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: coding | 1453 | 1469 | DeepSeek V4 Pro 0813 |
| LMArena Text: software and IT | 1449 | 1462 | DeepSeek V4 Pro 0813 |
| LMArena WebDev: React | 1350 | 1588 | DeepSeek V4 Pro 0813 |
| LMArena WebDev: HTML | 1371 | 1566 | DeepSeek V4 Pro 0813 |
| Neither model has a result on the other 13 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Pro 0813 | Leader |
|---|---|---|---|
| AA Intelligence Index | 21.5 | 36.0 | DeepSeek V4 Pro 0813 |
| Epoch Capabilities Index | 146.1 | 155.6 | DeepSeek V4 Pro 0813 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Pro 0813 | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 41.3 | Only DeepSeek V4 Pro 0813 measured |
| SREBench (Vals AI) | Not measured | 1.9% | Only DeepSeek V4 Pro 0813 measured |
| Finance Agent (Vals AI) | Not measured | 50.4% | Only DeepSeek V4 Pro 0813 measured |
| Tax Agent Bench (Vals AI) | Not measured | 26.8% | Only DeepSeek V4 Pro 0813 measured |
| Legal Research Bench (Vals AI) | Not measured | 40.9% | Only DeepSeek V4 Pro 0813 measured |
| Harvey's Legal Agent Benchmark (Vals AI) | Not measured | 10.8% | Only DeepSeek V4 Pro 0813 measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 52.8% | Only DeepSeek V4 Pro 0813 measured |
| Terminal-Bench-Science (Vals AI) | Not measured | 4.3% | Only DeepSeek V4 Pro 0813 measured |
| τ²-Bench Telecom | 90.6% | Not measured | Only DeepSeek V3.2 measured |
| APEX-Agents | 21.3% | 47.3% | DeepSeek V4 Pro 0813 |
| Vending-Bench 2 | $1,034 | Not measured | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 30 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index08
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Pro 0813 | Leader |
|---|---|---|---|
| GPQA Diamond | 80.3% | 92.4% | DeepSeek V4 Pro 0813 |
| GPQA Diamond (Artificial Analysis run) | 84.0% | 92.8% | DeepSeek V4 Pro 0813 |
| Humanity's Last Exam (Artificial Analysis run) | 24.6% | 41.0% | DeepSeek V4 Pro 0813 |
| GPQA Diamond (Epoch AI run) | Not measured | 91.7% | Only DeepSeek V4 Pro 0813 measured |
| ARC-AGI | 57.0% | 90.5% | DeepSeek V4 Pro 0813 |
| ARC-AGI-2 | 4.0% | 61.3% | DeepSeek V4 Pro 0813 |
| Chess puzzles | Not measured | 46.0% | Only DeepSeek V4 Pro 0813 measured |
| Mystery game puzzles | Not measured | 43.0% | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: hard prompts | 1434 | 1458 | DeepSeek V4 Pro 0813 |
| LMArena Text: expert | 1436 | 1459 | DeepSeek V4 Pro 0813 |
| LMArena Text: hard prompts in English | 1446 | 1463 | DeepSeek V4 Pro 0813 |
| Neither model has a result on the other 11 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Pro 0813 | Leader |
|---|---|---|---|
| AIME | 84.6% | Not measured | Only DeepSeek V3.2 measured |
| MGSM (Vals AI) | 86.0% | Not measured | Only DeepSeek V3.2 measured |
| AA Math Index | 92.0 | Not measured | Only DeepSeek V3.2 measured |
| AIME 2025 (Artificial Analysis run) | 92.0% | Not measured | Only DeepSeek V3.2 measured |
| OTIS Mock AIME | Not measured | 98.6% | Only DeepSeek V4 Pro 0813 measured |
| FrontierMath, tiers 1 to 3 | Not measured | 64.6% | Only DeepSeek V4 Pro 0813 measured |
| FrontierMath, tier 4 | Not measured | 26.8% | Only DeepSeek V4 Pro 0813 measured |
| ProofBench | Not measured | 50.0% | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: math | 1435 | 1441 | DeepSeek V4 Pro 0813 |
| LMArena Text: mathematical work | 1440 | 1443 | DeepSeek V4 Pro 0813 |
| Neither model has a result on the other 3 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index08
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Pro 0813 | Leader |
|---|---|---|---|
| MMLU Pro | 84.9% | 87.0% | DeepSeek V4 Pro 0813 |
| LegalBench (Vals AI) | 76.1% | 82.4% | DeepSeek V4 Pro 0813 |
| CaseLaw (Vals AI) | 55.4% | Not measured | Only DeepSeek V3.2 measured |
| CorpFin (Vals AI) | 51.0% | 65.4% | DeepSeek V4 Pro 0813 |
| TaxEval (Vals AI) | 68.2% | 73.1% | DeepSeek V4 Pro 0813 |
| MedQA (Vals AI) | 93.9% | Not measured | Only DeepSeek V3.2 measured |
| MedCode (Vals AI) | Not measured | 42.5% | Only DeepSeek V4 Pro 0813 measured |
| MMLU Pro (Artificial Analysis run) | 86.2% | Not measured | Only DeepSeek V3.2 measured |
| SimpleQA Verified | Not measured | 52.9% | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: science | 1446 | 1457 | DeepSeek V4 Pro 0813 |
| LMArena Text: medicine and healthcare | 1440 | 1446 | DeepSeek V4 Pro 0813 |
| LMArena Text: legal and government | 1425 | 1459 | DeepSeek V4 Pro 0813 |
| LMArena Text: business and finance | 1410 | 1442 | DeepSeek V4 Pro 0813 |
| Neither model has a result on the other 5 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Pro 0813 | Leader |
|---|---|---|---|
| Design Arena: websites | 1175 | Not measured | Only DeepSeek V3.2 measured |
| Design Arena: UI components | 1146 | Not measured | Only DeepSeek V3.2 measured |
| Design Arena: game dev | 1131 | Not measured | Only DeepSeek V3.2 measured |
| Design Arena: data viz | 1158 | Not measured | Only DeepSeek V3.2 measured |
| Design Arena: 3D | 1132 | Not measured | Only DeepSeek V3.2 measured |
| Design Arena: all code categories | 1163 | Not measured | Only DeepSeek V3.2 measured |
| Design Arena: SVG | 1032 | Not measured | Only DeepSeek V3.2 measured |
| Design Arena: ASCII art | 1083 | Not measured | 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 index017
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Pro 0813 | Leader |
|---|---|---|---|
| Vals Index | Not measured | 53.0% | Only DeepSeek V4 Pro 0813 measured |
| MedScribe (Vals AI) | Not measured | 80.2% | Only DeepSeek V4 Pro 0813 measured |
| IFBench | 60.7% | Not measured | Only DeepSeek V3.2 measured |
| Long context reasoning (AA-LCR) | 73.3% | 80.3% | DeepSeek V4 Pro 0813 |
| DTBench | 85.6% | 93.9% | DeepSeek V4 Pro 0813 |
| LMCA | 28.8% | 45.5% | DeepSeek V4 Pro 0813 |
| CL-bench | 12.4% | Not measured | Only DeepSeek V3.2 measured |
| CL-bench Life | 7.4% | Not measured | Only DeepSeek V3.2 measured |
| LMArena Text | 1425 | 1446 | DeepSeek V4 Pro 0813 |
| LMArena Text: instruction following | 1413 | 1448 | DeepSeek V4 Pro 0813 |
| LMArena Text: creative writing | 1400 | 1439 | DeepSeek V4 Pro 0813 |
| LMArena Text: English | 1439 | 1457 | DeepSeek V4 Pro 0813 |
| LMArena Text: non-English | 1409 | 1431 | DeepSeek V4 Pro 0813 |
| LMArena Text: Chinese | 1455 | 1481 | DeepSeek V4 Pro 0813 |
| LMArena Text: German | 1413 | Not measured | Only DeepSeek V3.2 measured |
| LMArena Text: Japanese | 1374 | Not measured | Only DeepSeek V3.2 measured |
| LMArena Text: Korean | 1370 | 1449 | DeepSeek V4 Pro 0813 |
| LMArena Text: French | 1429 | 1446 | DeepSeek V4 Pro 0813 |
| LMArena Text: Spanish | 1424 | 1435 | DeepSeek V4 Pro 0813 |
| LMArena Text: Russian | 1424 | 1448 | DeepSeek V4 Pro 0813 |
| LMArena Text: Polish | 1419 | Not measured | Only DeepSeek V3.2 measured |
| LMArena Text: multi-turn | 1427 | 1454 | DeepSeek V4 Pro 0813 |
| LMArena Text: longer queries | 1428 | 1458 | DeepSeek V4 Pro 0813 |
| LMArena Text: writing and language | 1407 | 1443 | DeepSeek V4 Pro 0813 |
| LMArena Text: entertainment and media | 1395 | 1424 | DeepSeek V4 Pro 0813 |
| Neither model has a result on the other 21 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Pro 0813 | 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 V3.2 | DeepSeek V4 Pro 0813 |
|---|---|---|
| Maker | DeepSeek | DeepSeek |
| Weights | Open weights | Open weights |
| Released | December 1, 2025 | August 12, 2026 |
| Context window | 164K | 1.05M |
| Input $/Mtok | $0.29 | $0.66 |
| Output $/Mtok | $0.88 | $1.98 |
| Coding index | 30.0 | 56.1 |
| Independent measurements behind it | 3 | 5 |
| Cost per solved task | $0.99 | $0.10 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V3.2 better than DeepSeek V4 Pro 0813 for coding?
Pick DeepSeek V4 Pro 0813 for coding: it scores 56.1 on the Vibe Coding Index against 30.0 for DeepSeek V3.2. Pick DeepSeek V3.2 for output tokens 2.3x cheaper ($0.88 vs $1.98 per million).
Which is cheaper, DeepSeek V3.2 or DeepSeek V4 Pro 0813?
DeepSeek V3.2 costs $0.29 per million input tokens and $0.88 per million output tokens; DeepSeek V4 Pro 0813 costs $0.66 and $1.98. Per solved SWE-bench Verified task, DeepSeek V3.2 costs $0.99 and DeepSeek V4 Pro 0813 $0.10.
Which has the bigger context window, DeepSeek V3.2 or DeepSeek V4 Pro 0813?
DeepSeek V4 Pro 0813, with 1.05M tokens against 164K.
How many benchmarks is the DeepSeek V3.2 vs DeepSeek V4 Pro 0813 comparison based on?
47 benchmarks have a result for both models as of October 22, 2026. DeepSeek V3.2's coding index rests on 3 independent benchmarks and DeepSeek V4 Pro 0813's on 5. Where only one model has a result, the page says so instead of naming a winner.