DeepSeek V3.2 Exp vs DeepSeek V4 Pro 0423
Updated October 21, 2026. 43 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick DeepSeek V3.2 Exp for coding: it scores 55.0 on the Vibe Coding Index against 35.3 for DeepSeek V4 Pro 0423.
Pick DeepSeek V4 Pro 0423 when the context window is larger, 1.05M against 164K tokens, or when what matters is LMArena WebDev or WeirdML, where it leads.
- Shared results
- 43
- DeepSeek V3.2 Exp only
- 14
- DeepSeek V4 Pro 0423 only
- 40
- Like-for-like categories
- 7 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
DeepSeek V3.2 Exp
DeepSeek V3.2 Exp scores 55.0 against 35.3.
2 and 1 benchmarks behind the scores
Intelligence
AA Intelligence Index
DeepSeek V4 Pro 0423
DeepSeek V4 Pro 0423 leads, 30.4 against 16.6.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only DeepSeek V4 Pro 0423 has a AA Agentic Index result, 26.3.
Directional only
Reasoning
GPQA Diamond
No call
Only DeepSeek V4 Pro 0423 has a GPQA Diamond result, 89.4%.
Directional only
Knowledge
MMLU Pro
No call
Only DeepSeek V4 Pro 0423 has a MMLU Pro result, 87.3%.
Directional only
Web design
Design Arena: websites
DeepSeek V4 Pro 0423
DeepSeek V4 Pro 0423 leads, 1235 against 1179.
Same board, both measured
Long documents
Listed context window
DeepSeek V4 Pro 0423
DeepSeek V4 Pro 0423 takes 1.05M tokens against 164K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
DeepSeek V3.2 Exp
DeepSeek V3.2 Exp $0.0005, DeepSeek V4 Pro 0423 $0.0019 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only DeepSeek V4 Pro 0423 has the number, $0.57.
Directional only
At a glance
- Coding index
- DeepSeek V3.2 Exp 55.0 from 2 benchmarks, DeepSeek V4 Pro 0423 35.3 from 1 benchmark.
- Widest category
- Coding, 14.1 points apart on average over 5 shared benchmarks.
- Widest single gap
- LMArena WebDev: HTML: DeepSeek V3.2 Exp 1281 against DeepSeek V4 Pro 0423 1435, 25.6 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V3.2 Exp $0.41 against DeepSeek V4 Pro 0423 $1.9105 per million output tokens, 4.7x apart.
- Context window
- DeepSeek V3.2 Exp takes 164K tokens, DeepSeek V4 Pro 0423 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 Exp | Bars | DeepSeek V4 Pro 0423 | Ties |
|---|---|---|---|---|
| Coding | 0 | 5 | ||
| Intelligence | 0 | 2 | ||
| Reasoning | 0 | 3 | ||
| Math | 0 | 2 | ||
| Knowledge | 1 | 3 | ||
| Web design | 0 | 7 | ||
| General | 0 | 20 | ||
| All shared boards | 1 | 42 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index05
| Benchmark | DeepSeek V3.2 Exp | DeepSeek V4 Pro 0423 | Leader |
|---|---|---|---|
| Aider polyglot | 74.2% | Not measured | Only DeepSeek V3.2 Exp measured |
| LMArena WebDev | 1272 | 1447 | DeepSeek V4 Pro 0423 |
| AA Coding Index | Not measured | 59.4 | Only DeepSeek V4 Pro 0423 measured |
| LiveCodeBench (Vals AI) | Not measured | 87.5% | Only DeepSeek V4 Pro 0423 measured |
| SWE-bench Verified (Vals AI) | Not measured | 77.4% | Only DeepSeek V4 Pro 0423 measured |
| Terminal-Bench 4 (Vals AI) | Not measured | 11.1% | Only DeepSeek V4 Pro 0423 measured |
| Vibe Code Bench (Vals AI) | Not measured | 49.9% | Only DeepSeek V4 Pro 0423 measured |
| Code migration (Vals AI) | Not measured | 26.2% | Only DeepSeek V4 Pro 0423 measured |
| ProgramBench (Vals AI) | Not measured | 0.0% | Only DeepSeek V4 Pro 0423 measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 50.2% | Only DeepSeek V4 Pro 0423 measured |
| SkillsBench (Vals AI) | Not measured | 51.3% | Only DeepSeek V4 Pro 0423 measured |
| LiveCodeBench (Artificial Analysis run) | 78.9% | Not measured | Only DeepSeek V3.2 Exp measured |
| Terminal-Bench Hard (Artificial Analysis run) | 31.1% | Not measured | Only DeepSeek V3.2 Exp measured |
| FrontierCode | Not measured | 17.6% | Only DeepSeek V4 Pro 0423 measured |
| WeirdML | 39.5% | 48.9% | DeepSeek V4 Pro 0423 |
| ALE-Bench | Not measured | 1006.1 | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: coding | 1438 | 1470 | DeepSeek V4 Pro 0423 |
| LMArena Text: software and IT | 1438 | 1469 | DeepSeek V4 Pro 0423 |
| LMArena WebDev: React | Not measured | 1441 | Only DeepSeek V4 Pro 0423 measured |
| LMArena WebDev: HTML | 1281 | 1435 | DeepSeek V4 Pro 0423 |
| Neither model has a result on the other 18 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V3.2 Exp | DeepSeek V4 Pro 0423 | Leader |
|---|---|---|---|
| AA Intelligence Index | 16.6 | 30.4 | DeepSeek V4 Pro 0423 |
| Epoch Capabilities Index | 144.9 | 149.1 | DeepSeek V4 Pro 0423 |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index11 boards, none shared
| Benchmark | DeepSeek V3.2 Exp | DeepSeek V4 Pro 0423 | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 26.3 | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: agents building web apps | Not measured | 1000 | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: agents building full-stack apps | Not measured | 948 | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: agents building Godot games | Not measured | 1059 | Only DeepSeek V4 Pro 0423 measured |
| Finance Agent (Vals AI) | Not measured | 44.1% | Only DeepSeek V4 Pro 0423 measured |
| Tax Agent Bench (Vals AI) | Not measured | 25.2% | Only DeepSeek V4 Pro 0423 measured |
| Legal Research Bench (Vals AI) | Not measured | 23.1% | Only DeepSeek V4 Pro 0423 measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 51.6% | Only DeepSeek V4 Pro 0423 measured |
| τ²-Bench Telecom | 33.9% | Not measured | Only DeepSeek V3.2 Exp measured |
| TheAgentCompany | 42.9% | Not measured | Only DeepSeek V3.2 Exp measured |
| Vending-Bench 2 | Not measured | $3,285 | Only DeepSeek V4 Pro 0423 measured |
| Neither model has a result on the other 28 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | DeepSeek V3.2 Exp | DeepSeek V4 Pro 0423 | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 89.4% | Only DeepSeek V4 Pro 0423 measured |
| GPQA Diamond (Artificial Analysis run) | 79.7% | Not measured | Only DeepSeek V3.2 Exp measured |
| Humanity's Last Exam (Artificial Analysis run) | 14.9% | Not measured | Only DeepSeek V3.2 Exp measured |
| GPQA Diamond (Epoch AI run) | Not measured | 90.9% | Only DeepSeek V4 Pro 0423 measured |
| Chess puzzles | Not measured | 19.0% | Only DeepSeek V4 Pro 0423 measured |
| Mystery game puzzles | Not measured | 17.0% | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: hard prompts | 1429 | 1461 | DeepSeek V4 Pro 0423 |
| LMArena Text: expert | 1422 | 1463 | DeepSeek V4 Pro 0423 |
| LMArena Text: hard prompts in English | 1444 | 1466 | DeepSeek V4 Pro 0423 |
| Neither model has a result on the other 12 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V3.2 Exp | DeepSeek V4 Pro 0423 | Leader |
|---|---|---|---|
| AA Math Index | 87.7 | Not measured | Only DeepSeek V3.2 Exp measured |
| AIME 2025 (Artificial Analysis run) | 87.7% | Not measured | Only DeepSeek V3.2 Exp measured |
| OTIS Mock AIME | Not measured | 96.7% | Only DeepSeek V4 Pro 0423 measured |
| FrontierMath, tiers 1 to 3 | Not measured | 45.3% | Only DeepSeek V4 Pro 0423 measured |
| FrontierMath, tier 4 | Not measured | 2.4% | Only DeepSeek V4 Pro 0423 measured |
| ProofBench | Not measured | 16.0% | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: math | 1422 | 1443 | DeepSeek V4 Pro 0423 |
| LMArena Text: mathematical work | 1433 | 1455 | DeepSeek V4 Pro 0423 |
| Neither model has a result on the other 5 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index13
| Benchmark | DeepSeek V3.2 Exp | DeepSeek V4 Pro 0423 | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 87.3% | Only DeepSeek V4 Pro 0423 measured |
| LegalBench (Vals AI) | Not measured | 80.3% | Only DeepSeek V4 Pro 0423 measured |
| CaseLaw (Vals AI) | Not measured | 59.4% | Only DeepSeek V4 Pro 0423 measured |
| CorpFin (Vals AI) | Not measured | 61.4% | Only DeepSeek V4 Pro 0423 measured |
| TaxEval (Vals AI) | Not measured | 72.1% | Only DeepSeek V4 Pro 0423 measured |
| MedCode (Vals AI) | Not measured | 40.5% | Only DeepSeek V4 Pro 0423 measured |
| MMLU Pro (Artificial Analysis run) | 85.0% | Not measured | Only DeepSeek V3.2 Exp measured |
| SimpleQA Verified | Not measured | 47.0% | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: science | 1443 | 1470 | DeepSeek V4 Pro 0423 |
| LMArena Text: medicine and healthcare | 1468 | 1466 | DeepSeek V3.2 Exp |
| LMArena Text: legal and government | 1419 | 1464 | DeepSeek V4 Pro 0423 |
| LMArena Text: business and finance | 1412 | 1443 | DeepSeek V4 Pro 0423 |
| Neither model has a result on the other 5 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index07
| Benchmark | DeepSeek V3.2 Exp | DeepSeek V4 Pro 0423 | Leader |
|---|---|---|---|
| Design Arena: websites | 1179 | 1235 | DeepSeek V4 Pro 0423 |
| Design Arena: UI components | 1166 | 1221 | DeepSeek V4 Pro 0423 |
| Design Arena: game dev | 1145 | 1229 | DeepSeek V4 Pro 0423 |
| Design Arena: data viz | 1152 | 1203 | DeepSeek V4 Pro 0423 |
| Design Arena: 3D | 1148 | 1254 | DeepSeek V4 Pro 0423 |
| Design Arena: all code categories | 1171 | 1240 | DeepSeek V4 Pro 0423 |
| Design Arena: SVG | 1031 | 1139 | DeepSeek V4 Pro 0423 |
| Design Arena: ASCII art | Not measured | 1148 | Only DeepSeek V4 Pro 0423 measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index020
| Benchmark | DeepSeek V3.2 Exp | DeepSeek V4 Pro 0423 | Leader |
|---|---|---|---|
| Public Benefits Bench (Vals AI) | Not measured | 62.9% | Only DeepSeek V4 Pro 0423 measured |
| MedScribe (Vals AI) | Not measured | 75.1% | Only DeepSeek V4 Pro 0423 measured |
| IFBench | 54.1% | Not measured | Only DeepSeek V3.2 Exp measured |
| Long context reasoning (AA-LCR) | 72.3% | Not measured | Only DeepSeek V3.2 Exp measured |
| DTBench | 87.7% | 90.7% | DeepSeek V4 Pro 0423 |
| LMCA | 29.1% | 41.2% | DeepSeek V4 Pro 0423 |
| Fiction.LiveBench | 83.3% | Not measured | Only DeepSeek V3.2 Exp measured |
| CL-bench | 13.2% | Not measured | Only DeepSeek V3.2 Exp measured |
| Surface Evolver Bench | Not measured | 40.0% | Only DeepSeek V4 Pro 0423 measured |
| CL-bench Life | 9.5% | 13.5% | DeepSeek V4 Pro 0423 |
| LMArena Text | 1425 | 1451 | DeepSeek V4 Pro 0423 |
| LMArena Text: instruction following | 1401 | 1445 | DeepSeek V4 Pro 0423 |
| LMArena Text: creative writing | 1404 | 1446 | DeepSeek V4 Pro 0423 |
| LMArena Text: English | 1441 | 1459 | DeepSeek V4 Pro 0423 |
| LMArena Text: non-English | 1404 | 1439 | DeepSeek V4 Pro 0423 |
| LMArena Text: Chinese | 1461 | 1481 | DeepSeek V4 Pro 0423 |
| LMArena Text: German | 1440 | 1458 | DeepSeek V4 Pro 0423 |
| LMArena Text: Japanese | 1331 | 1446 | DeepSeek V4 Pro 0423 |
| LMArena Text: Korean | 1372 | 1409 | DeepSeek V4 Pro 0423 |
| LMArena Text: French | 1434 | 1473 | DeepSeek V4 Pro 0423 |
| LMArena Text: Spanish | 1441 | 1457 | DeepSeek V4 Pro 0423 |
| LMArena Text: Russian | 1412 | 1453 | DeepSeek V4 Pro 0423 |
| LMArena Text: Polish | 1435 | 1461 | DeepSeek V4 Pro 0423 |
| LMArena Text: multi-turn | 1422 | 1468 | DeepSeek V4 Pro 0423 |
| LMArena Text: longer queries | 1418 | 1458 | DeepSeek V4 Pro 0423 |
| LMArena Text: writing and language | 1407 | 1447 | DeepSeek V4 Pro 0423 |
| LMArena Text: entertainment and media | 1407 | 1431 | DeepSeek V4 Pro 0423 |
| Neither model has a result on the other 19 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V3.2 Exp | DeepSeek V4 Pro 0423 | 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 Exp costs less
Repository review
50K input + 3K output tokens
DeepSeek V3.2 Exp costs less
Agent loop
200K input + 10K output tokens
DeepSeek V3.2 Exp 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 Exp | DeepSeek V4 Pro 0423 |
|---|---|---|
| Maker | DeepSeek | DeepSeek |
| Weights | Open weights | Open weights |
| Released | September 29, 2025 | April 24, 2026 |
| Context window | 164K | 1.05M |
| Input $/Mtok | $0.27 | $0.9553 |
| Output $/Mtok | $0.41 | $1.9105 |
| Coding index | 55.0 | 35.3 |
| Independent measurements behind it | 2 | 1 |
| Cost per solved task | Not computable | $0.57 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V3.2 Exp better than DeepSeek V4 Pro 0423 for coding?
Pick DeepSeek V3.2 Exp for coding: it scores 55.0 on the Vibe Coding Index against 35.3 for DeepSeek V4 Pro 0423. Pick DeepSeek V4 Pro 0423 when the context window is larger, 1.05M against 164K tokens, or when what matters is LMArena WebDev or WeirdML, where it leads.
Which is cheaper, DeepSeek V3.2 Exp or DeepSeek V4 Pro 0423?
DeepSeek V3.2 Exp costs $0.27 per million input tokens and $0.41 per million output tokens; DeepSeek V4 Pro 0423 costs $0.9553 and $1.9105. Cost per solved task is known for DeepSeek V4 Pro 0423 only.
Which has the bigger context window, DeepSeek V3.2 Exp or DeepSeek V4 Pro 0423?
DeepSeek V4 Pro 0423, with 1.05M tokens against 164K.
How many benchmarks is the DeepSeek V3.2 Exp vs DeepSeek V4 Pro 0423 comparison based on?
43 benchmarks have a result for both models as of October 21, 2026. DeepSeek V3.2 Exp's coding index rests on 2 independent benchmarks and DeepSeek V4 Pro 0423's on 1. Where only one model has a result, the page says so instead of naming a winner.