DeepSeek V3 vs o4 Mini
Updated October 21, 2026. 11 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 and o4 Mini are level for coding, 70.2 against 68.6 on the Vibe Coding Index, so price and context decide.
Pick o4 Mini when the context window is larger, 200K against 164K tokens, or when what matters is Aider polyglot or GPQA Diamond (Epoch AI run), where it leads. Pick DeepSeek V3 when output tokens cost 4.3x less, or when what matters is Design Arena: websites or Design Arena: UI components, where it leads.
- Shared results
- 11
- DeepSeek V3 only
- 2
- o4 Mini only
- 83
- Like-for-like categories
- 5 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
Level: 70.2 against 68.6, inside 2 points.
1 and 2 benchmarks behind the scores
Intelligence
AA Intelligence Index
No call
Only o4 Mini has a AA Intelligence Index result, 16.7.
Directional only
Reasoning
GPQA Diamond
No call
Only o4 Mini has a GPQA Diamond result, 74.5%.
Directional only
Math
AIME
No call
Only o4 Mini has a AIME result, 83.7%.
Directional only
Knowledge
MMLU Pro
No call
Only o4 Mini has a MMLU Pro result, 80.6%.
Directional only
Web design
Design Arena: websites
DeepSeek V3
DeepSeek V3 leads, 1120 against 985.
Same board, both measured
Long documents
Listed context window
o4 Mini
o4 Mini takes 200K tokens against 164K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
DeepSeek V3
DeepSeek V3 $0.0008, o4 Mini $0.0033 per turn.
List prices
At a glance
- Coding index
- DeepSeek V3 70.2 from 1 benchmark, o4 Mini 68.6 from 2 benchmarks.
- Widest category
- Math, 32.8 points apart on average over 1 shared benchmark.
- Widest single gap
- OTIS Mock AIME: DeepSeek V3 48.9% against o4 Mini 81.7%, 32.8 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V3 $1.0287 against o4 Mini $4.40 per million output tokens, 4.3x apart.
- Context window
- DeepSeek V3 takes 164K tokens, o4 Mini 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 | Bars | o4 Mini | Ties |
|---|---|---|---|---|
| Coding | 0 | 1 | ||
| Reasoning | 0 | 1 | ||
| Math | 0 | 1 | ||
| Web design | 6 | 0 | ||
| General | 0 | 2 | ||
| All shared boards | 6 | 5 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index01
| Benchmark | DeepSeek V3 | o4 Mini | Leader |
|---|---|---|---|
| Aider polyglot | 70.2% | 72.0% | o4 Mini |
| SWE-bench Verified | Not measured | 64.6% | Only o4 Mini measured |
| SWE-bench Lite | 30.7% | Not measured | Only DeepSeek V3 measured |
| SWE-bench Multimodal | Not measured | 33.9% | Only o4 Mini measured |
| LiveCodeBench (Vals AI) | Not measured | 82.2% | Only o4 Mini measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 85.9% | Only o4 Mini measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 15.2% | Only o4 Mini measured |
| GSO-Bench | Not measured | 3.6% | Only o4 Mini measured |
| WeirdML | Not measured | 52.6% | Only o4 Mini measured |
| LMArena Text: coding | Not measured | 1368 | Only o4 Mini measured |
| LMArena Text: software and IT | Not measured | 1367 | Only o4 Mini measured |
| Neither model has a result on the other 27 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | DeepSeek V3 | o4 Mini | Leader |
|---|---|---|---|
| AA Intelligence Index | Not measured | 16.7 | Only o4 Mini measured |
| Epoch Capabilities Index | Not measured | 145.5 | Only o4 Mini measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | DeepSeek V3 | o4 Mini | Leader |
|---|---|---|---|
| τ²-Bench Telecom | Not measured | 55.6% | Only o4 Mini measured |
| METR time horizons | Not measured | 63.9% | Only o4 Mini measured |
| Neither model has a result on the other 37 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | DeepSeek V3 | o4 Mini | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 74.5% | Only o4 Mini measured |
| MMMU Pro (Vals AI) | Not measured | 79.7% | Only o4 Mini measured |
| GPQA Diamond (Artificial Analysis run) | Not measured | 78.4% | Only o4 Mini measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 16.5% | Only o4 Mini measured |
| GPQA Diamond (Epoch AI run) | 71.2% | 79.6% | o4 Mini |
| ARC-AGI | Not measured | 58.7% | Only o4 Mini measured |
| ARC-AGI-2 | Not measured | 6.1% | Only o4 Mini measured |
| SimpleBench | Not measured | 38.7% | Only o4 Mini measured |
| Chess puzzles | Not measured | 22.0% | Only o4 Mini measured |
| Mystery game puzzles | Not measured | 5.0% | Only o4 Mini measured |
| VPCT | Not measured | 57.5% | Only o4 Mini measured |
| LMArena Text: hard prompts | Not measured | 1351 | Only o4 Mini measured |
| LMArena Text: expert | Not measured | 1343 | Only o4 Mini measured |
| LMArena Text: hard prompts in English | Not measured | 1366 | Only o4 Mini measured |
| LMArena Vision: diagrams | Not measured | 1197 | Only o4 Mini measured |
| LMArena Vision: homework | Not measured | 1234 | Only o4 Mini measured |
| Neither model has a result on the other 5 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | DeepSeek V3 | o4 Mini | Leader |
|---|---|---|---|
| AIME | Not measured | 83.7% | Only o4 Mini measured |
| MATH 500 (Vals AI) | Not measured | 94.2% | Only o4 Mini measured |
| MGSM (Vals AI) | Not measured | 93.4% | Only o4 Mini measured |
| AA Math Index | Not measured | 90.7 | Only o4 Mini measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 90.7% | Only o4 Mini measured |
| OTIS Mock AIME | 48.9% | 81.7% | o4 Mini |
| FrontierMath, tiers 1 to 3 | Not measured | 36.1% | Only o4 Mini measured |
| FrontierMath, tier 4 | Not measured | 4.9% | Only o4 Mini measured |
| MATH level 5 | Not measured | 97.8% | Only o4 Mini measured |
| LMArena Text: math | Not measured | 1389 | Only o4 Mini measured |
| LMArena Text: mathematical work | Not measured | 1379 | Only o4 Mini measured |
| Neither model has a result on the other 2 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index14 boards, none shared
| Benchmark | DeepSeek V3 | o4 Mini | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 80.6% | Only o4 Mini measured |
| LegalBench (Vals AI) | Not measured | 79.2% | Only o4 Mini measured |
| CorpFin (Vals AI) | Not measured | 59.0% | Only o4 Mini measured |
| TaxEval (Vals AI) | Not measured | 74.8% | Only o4 Mini measured |
| MortgageTax (Vals AI) | Not measured | 64.8% | Only o4 Mini measured |
| MedQA (Vals AI) | Not measured | 96.0% | Only o4 Mini measured |
| MedCode (Vals AI) | Not measured | 33.8% | Only o4 Mini measured |
| MMLU Pro (Artificial Analysis run) | Not measured | 83.2% | Only o4 Mini measured |
| SimpleQA Verified | Not measured | 19.6% | Only o4 Mini measured |
| GeoBench | Not measured | 64.0% | Only o4 Mini measured |
| LMArena Text: science | Not measured | 1354 | Only o4 Mini measured |
| LMArena Text: medicine and healthcare | Not measured | 1361 | Only o4 Mini measured |
| LMArena Text: legal and government | Not measured | 1350 | Only o4 Mini measured |
| LMArena Text: business and finance | Not measured | 1325 | Only o4 Mini measured |
| Neither model has a result on the other 3 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | DeepSeek V3 | o4 Mini | Leader |
|---|---|---|---|
| Design Arena: websites | 1120 | 985 | DeepSeek V3 |
| Design Arena: UI components | 1093 | 983 | DeepSeek V3 |
| Design Arena: game dev | 1054 | 1004 | DeepSeek V3 |
| Design Arena: data viz | 1089 | 990 | DeepSeek V3 |
| Design Arena: 3D | 1087 | 854 | DeepSeek V3 |
| Design Arena: all code categories | 1108 | 977 | DeepSeek V3 |
| Design Arena: SVG | 976 | Not measured | Only DeepSeek V3 measured |
| 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 | o4 Mini | Leader |
|---|---|---|---|
| SAGE (Vals AI) | Not measured | 41.1% | Only o4 Mini measured |
| MedScribe (Vals AI) | Not measured | 69.1% | Only o4 Mini measured |
| IFBench | Not measured | 68.7% | Only o4 Mini measured |
| Long context reasoning (AA-LCR) | Not measured | 61.0% | Only o4 Mini measured |
| DTBench | 62.7% | 77.6% | o4 Mini |
| LMCA | 15.2% | 26.5% | o4 Mini |
| Fiction.LiveBench | Not measured | 77.8% | Only o4 Mini measured |
| Creative writing (Lech Mazur) | Not measured | 75.0% | Only o4 Mini measured |
| LMArena Text | Not measured | 1353 | Only o4 Mini measured |
| LMArena Text: instruction following | Not measured | 1320 | Only o4 Mini measured |
| LMArena Text: creative writing | Not measured | 1294 | Only o4 Mini measured |
| LMArena Vision | Not measured | 1194 | Only o4 Mini measured |
| LMArena Text: English | Not measured | 1367 | Only o4 Mini measured |
| LMArena Text: non-English | Not measured | 1337 | Only o4 Mini measured |
| LMArena Text: Chinese | Not measured | 1354 | Only o4 Mini measured |
| LMArena Text: German | Not measured | 1336 | Only o4 Mini measured |
| LMArena Text: Japanese | Not measured | 1308 | Only o4 Mini measured |
| LMArena Text: Korean | Not measured | 1312 | Only o4 Mini measured |
| LMArena Text: French | Not measured | 1364 | Only o4 Mini measured |
| LMArena Text: Spanish | Not measured | 1346 | Only o4 Mini measured |
| LMArena Text: Russian | Not measured | 1334 | Only o4 Mini measured |
| LMArena Text: Polish | Not measured | 1358 | Only o4 Mini measured |
| LMArena Text: multi-turn | Not measured | 1350 | Only o4 Mini measured |
| LMArena Text: longer queries | Not measured | 1315 | Only o4 Mini measured |
| LMArena Text: writing and language | Not measured | 1316 | Only o4 Mini measured |
| LMArena Text: entertainment and media | Not measured | 1308 | Only o4 Mini measured |
| LMArena Vision: English | Not measured | 1200 | Only o4 Mini measured |
| LMArena Vision: Chinese | Not measured | 1218 | Only o4 Mini measured |
| LMArena Vision: reading text | Not measured | 1197 | Only o4 Mini measured |
| LMArena Vision: humor | Not measured | 1199 | Only o4 Mini measured |
| LMArena Vision: captioning | Not measured | 1204 | Only o4 Mini measured |
| LMArena Vision: recognizing things | Not measured | 1218 | Only o4 Mini measured |
| LMArena Vision: creative writing | Not measured | 1166 | Only o4 Mini measured |
| Neither model has a result on the other 13 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V3 | o4 Mini | 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 costs less
Repository review
50K input + 3K output tokens
DeepSeek V3 costs less
Agent loop
200K input + 10K output tokens
DeepSeek V3 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 | OpenAI o4 Mini |
|---|---|---|
| Maker | DeepSeek | OpenAI |
| Weights | Open weights | No public weights listed |
| Released | December 26, 2024 | April 16, 2025 |
| Context window | 164K | 200K |
| Input $/Mtok | $0.2574 | $1.10 |
| Output $/Mtok | $1.0287 | $4.40 |
| Coding index | 70.2 | 68.6 |
| Independent measurements behind it | 1 | 2 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V3 better than o4 Mini for coding?
DeepSeek V3 and o4 Mini are level for coding, 70.2 against 68.6 on the Vibe Coding Index, so price and context decide. Pick o4 Mini when the context window is larger, 200K against 164K tokens, or when what matters is Aider polyglot or GPQA Diamond (Epoch AI run), where it leads. Pick DeepSeek V3 when output tokens cost 4.3x less, or when what matters is Design Arena: websites or Design Arena: UI components, where it leads.
Which is cheaper, DeepSeek V3 or o4 Mini?
DeepSeek V3 costs $0.2574 per million input tokens and $1.0287 per million output tokens; o4 Mini costs $1.10 and $4.40. Cost per solved task is known for neither of them.
Which has the bigger context window, DeepSeek V3 or o4 Mini?
o4 Mini, with 200K tokens against 164K.
How many benchmarks is the DeepSeek V3 vs o4 Mini comparison based on?
11 benchmarks have a result for both models as of October 21, 2026. DeepSeek V3's coding index rests on 1 independent benchmark and o4 Mini's on 2. Where only one model has a result, the page says so instead of naming a winner.