DeepSeek V3 vs DeepSeek V3.2 Exp
DeepSeek V3.2 Exp costs 2.3x less per output token than DeepSeek V3 ($0.41 vs $0.9594 per million). Updated October 22, 2026; 11 benchmarks have a result for both.
DeepSeek V3 or DeepSeek V3.2 Exp: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick DeepSeek V3 for work measured by Aider polyglot or Design Arena: websites, where it leads.
- Shared results
- 11
- DeepSeek V3 only
- 3
- DeepSeek V3.2 Exp only
- 48
- Like-for-like categories
- 4 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
No call
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
No call
Only DeepSeek V3.2 Exp has a AA Intelligence Index result, 16.6.
Directional only
Web design
Design Arena: websites
DeepSeek V3.2 Exp
DeepSeek V3.2 Exp leads, 1178 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 Exp
DeepSeek V3 $0.0008, DeepSeek V3.2 Exp $0.0005 per turn.
List prices
At a glance
- Coding index
- DeepSeek V3 not benchmarked, DeepSeek V3.2 Exp not benchmarked.
- Widest category
- General, 19.5 points apart on average over 2 shared benchmarks.
- Widest single gap
- DTBench: DeepSeek V3 62.7% against DeepSeek V3.2 Exp 87.7%, 25.1 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V3 $0.9594 against DeepSeek V3.2 Exp $0.41 per million output tokens, 2.3x 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 Exp | Ties |
|---|---|---|---|---|
| Coding | 0 | 1 | ||
| Reasoning | 1 | 0 | ||
| Web design | 0 | 7 | ||
| General | 0 | 2 | ||
| All shared boards | 1 | 10 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index01
| Benchmark | DeepSeek V3 | DeepSeek V3.2 Exp | Leader |
|---|---|---|---|
| Aider polyglot | 70.2% | 74.2% | DeepSeek V3.2 Exp |
| SWE-bench Lite | 30.7% | Not measured | Only DeepSeek V3 measured |
| LMArena WebDev | Not measured | 1272 | Only DeepSeek V3.2 Exp measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 78.9% | Only DeepSeek V3.2 Exp measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 31.1% | Only DeepSeek V3.2 Exp measured |
| WeirdML | Not measured | 39.5% | Only DeepSeek V3.2 Exp measured |
| LMArena Text: coding | Not measured | 1438 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: software and IT | Not measured | 1438 | Only DeepSeek V3.2 Exp measured |
| LMArena WebDev: HTML | Not measured | 1281 | Only DeepSeek V3.2 Exp measured |
| Neither model has a result on the other 30 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 Exp | Leader |
|---|---|---|---|
| AA Intelligence Index | Not measured | 16.6 | Only DeepSeek V3.2 Exp measured |
| Epoch Capabilities Index | Not measured | 144.9 | Only DeepSeek V3.2 Exp measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index3 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 Exp | Leader |
|---|---|---|---|
| τ²-Bench Telecom | Not measured | 33.9% | Only DeepSeek V3.2 Exp measured |
| TheAgentCompany | Not measured | 42.9% | Only DeepSeek V3.2 Exp measured |
| Berkeley Function Calling Leaderboard (BFCL) | Not measured | 56.7% | Only DeepSeek V3.2 Exp measured |
| Neither model has a result on the other 38 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | DeepSeek V3 | DeepSeek V3.2 Exp | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | Not measured | 79.7% | Only DeepSeek V3.2 Exp measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 14.9% | Only DeepSeek V3.2 Exp measured |
| GPQA Diamond (Epoch AI run) | 71.2% | Not measured | Only DeepSeek V3 measured |
| LMArena Text: hard prompts | Not measured | 1429 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: expert | Not measured | 1422 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: hard prompts in English | Not measured | 1444 | Only DeepSeek V3.2 Exp measured |
| Kagi LLM Benchmark | 52.3% | 52.2% | DeepSeek V3 |
| Neither model has a result on the other 15 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 Exp | Leader |
|---|---|---|---|
| AA Math Index | Not measured | 87.7 | Only DeepSeek V3.2 Exp measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 87.7% | Only DeepSeek V3.2 Exp measured |
| OTIS Mock AIME | 48.9% | Not measured | Only DeepSeek V3 measured |
| LMArena Text: math | Not measured | 1422 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: mathematical work | Not measured | 1433 | Only DeepSeek V3.2 Exp measured |
| Neither model has a result on the other 8 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 Exp | Leader |
|---|---|---|---|
| MMLU Pro (Artificial Analysis run) | Not measured | 85.0% | Only DeepSeek V3.2 Exp measured |
| LMArena Text: science | Not measured | 1443 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: medicine and healthcare | Not measured | 1468 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: legal and government | Not measured | 1419 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: business and finance | Not measured | 1412 | Only DeepSeek V3.2 Exp measured |
| Neither model has a result on the other 13 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index07
| Benchmark | DeepSeek V3 | DeepSeek V3.2 Exp | Leader |
|---|---|---|---|
| Design Arena: websites | 1120 | 1178 | DeepSeek V3.2 Exp |
| Design Arena: UI components | 1090 | 1164 | DeepSeek V3.2 Exp |
| Design Arena: game dev | 1053 | 1144 | DeepSeek V3.2 Exp |
| Design Arena: data viz | 1089 | 1152 | DeepSeek V3.2 Exp |
| Design Arena: 3D | 1086 | 1148 | DeepSeek V3.2 Exp |
| Design Arena: all code categories | 1108 | 1170 | DeepSeek V3.2 Exp |
| Design Arena: SVG | 976 | 1031 | DeepSeek V3.2 Exp |
| Neither model has a result on the other 2 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V3 | DeepSeek V3.2 Exp | Leader |
|---|---|---|---|
| IFBench | Not measured | 54.1% | Only DeepSeek V3.2 Exp measured |
| Long context reasoning (AA-LCR) | Not measured | 72.3% | Only DeepSeek V3.2 Exp measured |
| DTBench | 62.7% | 87.7% | DeepSeek V3.2 Exp |
| LMCA | 15.2% | 29.1% | DeepSeek V3.2 Exp |
| Fiction.LiveBench | Not measured | 83.3% | Only DeepSeek V3.2 Exp measured |
| CL-bench | Not measured | 13.2% | Only DeepSeek V3.2 Exp measured |
| CL-bench Life | Not measured | 9.5% | Only DeepSeek V3.2 Exp measured |
| LMArena Text | Not measured | 1425 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: instruction following | Not measured | 1401 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: creative writing | Not measured | 1404 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: English | Not measured | 1441 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: non-English | Not measured | 1404 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: Chinese | Not measured | 1461 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: German | Not measured | 1440 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: Japanese | Not measured | 1331 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: Korean | Not measured | 1372 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: French | Not measured | 1434 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: Spanish | Not measured | 1441 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: Russian | Not measured | 1412 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: Polish | Not measured | 1435 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: multi-turn | Not measured | 1422 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: longer queries | Not measured | 1418 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: writing and language | Not measured | 1407 | Only DeepSeek V3.2 Exp measured |
| LMArena Text: entertainment and media | Not measured | 1407 | Only DeepSeek V3.2 Exp measured |
| Neither model has a result on the other 22 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V3 | DeepSeek V3.2 Exp | 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 | DeepSeek V3.2 Exp |
|---|---|---|
| Maker | DeepSeek | DeepSeek |
| Weights | Open weights | Open weights |
| Released | December 26, 2024 | September 29, 2025 |
| Context window | 164K | 164K |
| Input $/Mtok | $0.2887 | $0.27 |
| Output $/Mtok | $0.9594 | $0.41 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 1 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V3 better than DeepSeek V3.2 Exp for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick DeepSeek V3 for work measured by Aider polyglot or Design Arena: websites, where it leads.
Which is cheaper, DeepSeek V3 or DeepSeek V3.2 Exp?
DeepSeek V3 costs $0.2887 per million input tokens and $0.9594 per million output tokens; DeepSeek V3.2 Exp costs $0.27 and $0.41. Cost per solved task is known for neither of them.
Which has the bigger context window, DeepSeek V3 or DeepSeek V3.2 Exp?
Neither: both take 164K tokens.
How many benchmarks is the DeepSeek V3 vs DeepSeek V3.2 Exp comparison based on?
11 benchmarks have a result for both models as of October 22, 2026. DeepSeek V3's coding index rests on 0 independent benchmarks and DeepSeek V3.2 Exp's on 1. Where only one model has a result, the page says so instead of naming a winner.