DeepSeek V3.2 vs o3
Updated October 21, 2026. 60 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick o3 for coding: it scores 70.7 on the Vibe Coding Index against 54.4 for DeepSeek V3.2.
Pick DeepSeek V3.2 when output tokens cost 19.0x less, or when what matters is SWE-bench Verified or LiveCodeBench (Artificial Analysis run), where it leads.
- Shared results
- 60
- DeepSeek V3.2 only
- 15
- o3 only
- 37
- Like-for-like categories
- 8 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
o3
o3 scores 70.7 against 54.4.
2 benchmarks behind each score
Intelligence
AA Intelligence Index
DeepSeek V3.2
DeepSeek V3.2 leads, 21.5 against 20.2.
Same board, both measured
Reasoning
GPQA Diamond
o3
o3 leads, 84.1% against 80.3%.
Same board, both measured
Math
AIME
o3
o3 leads, 85.3% against 84.6%.
Same board, both measured
Knowledge
MMLU Pro
o3
o3 leads, 85.6% against 84.9%.
Same board, both measured
Web design
Design Arena: websites
DeepSeek V3.2
DeepSeek V3.2 leads, 1175 against 1036.
Same board, both measured
Long documents
Listed context window
o3
o3 takes 200K tokens against 164K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
DeepSeek V3.2
DeepSeek V3.2 $0.0005, o3 $0.0060 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
o3
DeepSeek V3.2 $0.99, o3 $0.571.
Published attempt cost and pass rate
At a glance
- Coding index
- DeepSeek V3.2 54.4 from 2 benchmarks, o3 70.7 from 2 benchmarks.
- Widest category
- Web design, 13.9 points apart on average over 5 shared benchmarks.
- Widest single gap
- Design Arena: all code categories: DeepSeek V3.2 1164 against o3 1021, 17.9 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V3.2 $0.42 against o3 $8.00 per million output tokens, 19.0x apart.
- Context window
- DeepSeek V3.2 takes 164K tokens, o3 200K.
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 | o3 | Ties |
|---|---|---|---|---|
| Coding | 4 | 2 | ||
| Intelligence | 1 | 1 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 5 | 3 | ||
| Math | 4 | 2 | ||
| Knowledge | 3 | 7 | ||
| Web design | 4 | 1 | ||
| General | 14 | 8 | ||
| All shared boards | 36 | 24 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index42
| Benchmark | DeepSeek V3.2 | o3 | Leader |
|---|---|---|---|
| Aider polyglot | Not measured | 81.3% | Only o3 measured |
| SWE-bench Verified | 70.0% | 58.4% | DeepSeek V3.2 |
| SWE-bench Multimodal | Not measured | 36.0% | Only o3 measured |
| SWE-bench Multilingual | 59.0% | Not measured | Only DeepSeek V3.2 measured |
| LMArena WebDev | 1362 | Not measured | Only DeepSeek V3.2 measured |
| AA Coding Index | 44.2 | Not measured | Only DeepSeek V3.2 measured |
| LiveCodeBench (Vals AI) | 80.7% | 83.9% | o3 |
| SWE-bench Verified (Vals AI) | 67.6% | Not measured | Only DeepSeek V3.2 measured |
| Vibe Code Bench (Vals AI) | 5.1% | Not measured | Only DeepSeek V3.2 measured |
| LiveCodeBench (Artificial Analysis run) | 86.2% | 80.8% | DeepSeek V3.2 |
| Terminal-Bench Hard (Artificial Analysis run) | 35.6% | 37.1% | o3 |
| Terminal-Bench | 39.6% | Not measured | Only DeepSeek V3.2 measured |
| GSO-Bench | Not measured | 8.8% | Only o3 measured |
| WeirdML | Not measured | 52.4% | Only o3 measured |
| ALE-Bench | Not measured | 933.5 | Only o3 measured |
| LMArena Text: coding | 1453 | 1407 | DeepSeek V3.2 |
| LMArena Text: software and IT | 1449 | 1415 | DeepSeek V3.2 |
| LMArena WebDev: React | 1350 | Not measured | Only DeepSeek V3.2 measured |
| LMArena WebDev: HTML | 1371 | Not measured | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 19 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index11
| Benchmark | DeepSeek V3.2 | o3 | Leader |
|---|---|---|---|
| AA Intelligence Index | 21.5 | 20.2 | DeepSeek V3.2 |
| Epoch Capabilities Index | 146.1 | 146.7 | o3 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | DeepSeek V3.2 | o3 | Leader |
|---|---|---|---|
| τ²-Bench Telecom | 90.6% | 80.7% | DeepSeek V3.2 |
| APEX-Agents | 21.3% | Not measured | Only DeepSeek V3.2 measured |
| METR time horizons | Not measured | 65.4% | Only o3 measured |
| GDPval | Not measured | 30.8% | Only o3 measured |
| DeepResearch Bench | Not measured | 45.2% | Only o3 measured |
| OSWorld | Not measured | 23.0% | Only o3 measured |
| Vending-Bench 2 | $1,034 | Not measured | Only DeepSeek V3.2 measured |
| Neither model has a result on the other 32 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index53
| Benchmark | DeepSeek V3.2 | o3 | Leader |
|---|---|---|---|
| GPQA Diamond | 80.3% | 84.1% | o3 |
| MMMU Pro (Vals AI) | Not measured | 80.4% | Only o3 measured |
| GPQA Diamond (Artificial Analysis run) | 84.0% | 82.7% | DeepSeek V3.2 |
| Humanity's Last Exam (Artificial Analysis run) | 24.6% | 20.1% | DeepSeek V3.2 |
| GPQA Diamond (Epoch AI run) | Not measured | 81.8% | Only o3 measured |
| ARC-AGI | 57.0% | 60.8% | o3 |
| ARC-AGI-2 | 4.0% | 6.5% | o3 |
| SimpleBench | Not measured | 53.1% | Only o3 measured |
| Chess puzzles | Not measured | 37.0% | Only o3 measured |
| Mystery game puzzles | Not measured | 29.0% | Only o3 measured |
| VPCT | Not measured | 52.0% | Only o3 measured |
| LMArena Text: hard prompts | 1434 | 1402 | DeepSeek V3.2 |
| LMArena Text: expert | 1436 | 1402 | DeepSeek V3.2 |
| LMArena Text: hard prompts in English | 1446 | 1411 | DeepSeek V3.2 |
| LMArena Vision: diagrams | Not measured | 1220 | Only o3 measured |
| LMArena Vision: homework | Not measured | 1234 | Only o3 measured |
| Neither model has a result on the other 5 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index42
| Benchmark | DeepSeek V3.2 | o3 | Leader |
|---|---|---|---|
| AIME | 84.6% | 85.3% | o3 |
| MATH 500 (Vals AI) | Not measured | 94.6% | Only o3 measured |
| MGSM (Vals AI) | 86.0% | 91.8% | o3 |
| AA Math Index | 92.0 | 88.3 | DeepSeek V3.2 |
| AIME 2025 (Artificial Analysis run) | 92.0% | 88.3% | DeepSeek V3.2 |
| OTIS Mock AIME | Not measured | 84.4% | Only o3 measured |
| FrontierMath, tiers 1 to 3 | Not measured | 33.3% | Only o3 measured |
| MATH level 5 | Not measured | 97.8% | Only o3 measured |
| LMArena Text: math | 1435 | 1426 | DeepSeek V3.2 |
| LMArena Text: mathematical work | 1440 | 1423 | DeepSeek V3.2 |
| Neither model has a result on the other 3 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index37
| Benchmark | DeepSeek V3.2 | o3 | Leader |
|---|---|---|---|
| MMLU Pro | 84.9% | 85.6% | o3 |
| LegalBench (Vals AI) | 76.1% | 83.8% | o3 |
| CaseLaw (Vals AI) | 55.4% | Not measured | Only DeepSeek V3.2 measured |
| CorpFin (Vals AI) | 51.0% | 59.7% | o3 |
| TaxEval (Vals AI) | 68.2% | 74.6% | o3 |
| MortgageTax (Vals AI) | Not measured | 65.7% | Only o3 measured |
| MedQA (Vals AI) | 93.9% | 96.1% | o3 |
| MedCode (Vals AI) | Not measured | 47.3% | Only o3 measured |
| MMLU Pro (Artificial Analysis run) | 86.2% | 85.3% | DeepSeek V3.2 |
| SimpleQA Verified | Not measured | 49.4% | Only o3 measured |
| GeoBench | Not measured | 74.0% | Only o3 measured |
| LMArena Text: science | 1446 | 1429 | DeepSeek V3.2 |
| LMArena Text: medicine and healthcare | 1440 | 1448 | o3 |
| LMArena Text: legal and government | 1425 | 1428 | o3 |
| LMArena Text: business and finance | 1410 | 1393 | DeepSeek V3.2 |
| Neither model has a result on the other 2 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index41
| Benchmark | DeepSeek V3.2 | o3 | Leader |
|---|---|---|---|
| Design Arena: websites | 1175 | 1036 | DeepSeek V3.2 |
| Design Arena: UI components | 1149 | 1016 | DeepSeek V3.2 |
| Design Arena: game dev | 1132 | 1034 | DeepSeek V3.2 |
| Design Arena: data viz | 1158 | 1200 | o3 |
| Design Arena: 3D | 1133 | Not measured | Only DeepSeek V3.2 measured |
| Design Arena: all code categories | 1164 | 1021 | DeepSeek V3.2 |
| Design Arena: SVG | 1032 | Not measured | Only DeepSeek V3.2 measured |
| Design Arena: ASCII art | 1082 | 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 index148
| Benchmark | DeepSeek V3.2 | o3 | Leader |
|---|---|---|---|
| SAGE (Vals AI) | Not measured | 41.8% | Only o3 measured |
| MedScribe (Vals AI) | Not measured | 76.7% | Only o3 measured |
| IFBench | 60.7% | 71.4% | o3 |
| Long context reasoning (AA-LCR) | 73.3% | 74.7% | o3 |
| DTBench | 85.6% | 84.8% | DeepSeek V3.2 |
| LMCA | 28.8% | 39.7% | o3 |
| Fiction.LiveBench | Not measured | 88.9% | Only o3 measured |
| Creative writing (Lech Mazur) | Not measured | 83.9% | Only o3 measured |
| CL-bench | 12.4% | 17.8% | o3 |
| CL-bench Life | 7.4% | Not measured | Only DeepSeek V3.2 measured |
| LMArena Text | 1425 | 1410 | DeepSeek V3.2 |
| LMArena Text: instruction following | 1413 | 1368 | DeepSeek V3.2 |
| LMArena Text: creative writing | 1400 | 1359 | DeepSeek V3.2 |
| LMArena Vision | Not measured | 1214 | Only o3 measured |
| LMArena Text: English | 1439 | 1417 | DeepSeek V3.2 |
| LMArena Text: non-English | 1409 | 1401 | DeepSeek V3.2 |
| LMArena Text: Chinese | 1455 | 1437 | DeepSeek V3.2 |
| LMArena Text: German | 1413 | 1420 | o3 |
| LMArena Text: Japanese | 1374 | 1403 | o3 |
| LMArena Text: Korean | 1370 | 1370 | DeepSeek V3.2 |
| LMArena Text: French | 1429 | 1430 | o3 |
| LMArena Text: Spanish | 1424 | 1395 | DeepSeek V3.2 |
| LMArena Text: Russian | 1424 | 1406 | DeepSeek V3.2 |
| LMArena Text: Polish | 1419 | 1438 | o3 |
| LMArena Text: multi-turn | 1427 | 1405 | DeepSeek V3.2 |
| LMArena Text: longer queries | 1428 | 1372 | DeepSeek V3.2 |
| LMArena Text: writing and language | 1407 | 1372 | DeepSeek V3.2 |
| LMArena Text: entertainment and media | 1395 | 1373 | DeepSeek V3.2 |
| LMArena Vision: English | Not measured | 1221 | Only o3 measured |
| LMArena Vision: Chinese | Not measured | 1237 | Only o3 measured |
| LMArena Vision: reading text | Not measured | 1218 | Only o3 measured |
| LMArena Vision: humor | Not measured | 1222 | Only o3 measured |
| LMArena Vision: captioning | Not measured | 1210 | Only o3 measured |
| LMArena Vision: recognizing things | Not measured | 1234 | Only o3 measured |
| LMArena Vision: creative writing | Not measured | 1201 | Only o3 measured |
| Neither model has a result on the other 11 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V3.2 | o3 | 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 | OpenAI o3 |
|---|---|---|
| Maker | DeepSeek | OpenAI |
| Weights | Open weights | No public weights listed |
| Released | December 1, 2025 | April 16, 2025 |
| Context window | 164K | 200K |
| Input $/Mtok | $0.28 | $2.00 |
| Output $/Mtok | $0.42 | $8.00 |
| Coding index | 54.4 | 70.7 |
| Independent measurements behind it | 2 | 2 |
| Cost per solved task | $0.99 | $0.571 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V3.2 better than o3 for coding?
Pick o3 for coding: it scores 70.7 on the Vibe Coding Index against 54.4 for DeepSeek V3.2. Pick DeepSeek V3.2 when output tokens cost 19.0x less, or when what matters is SWE-bench Verified or LiveCodeBench (Artificial Analysis run), where it leads.
Which is cheaper, DeepSeek V3.2 or o3?
DeepSeek V3.2 costs $0.28 per million input tokens and $0.42 per million output tokens; o3 costs $2.00 and $8.00. Per solved SWE-bench Verified task, DeepSeek V3.2 costs $0.99 and o3 $0.571.
Which has the bigger context window, DeepSeek V3.2 or o3?
o3, with 200K tokens against 164K.
How many benchmarks is the DeepSeek V3.2 vs o3 comparison based on?
60 benchmarks have a result for both models as of October 21, 2026. DeepSeek V3.2's coding index rests on 2 independent benchmarks and o3's on 2. Where only one model has a result, the page says so instead of naming a winner.