DeepSeek V4 Pro 0423 vs Gemma 3 12B
DeepSeek V4 Pro 0423 leads Gemma 3 12B on the Artificial Analysis Intelligence Index, 30.4 to 3.8, and Gemma 3 12B costs 16.7x less per output token ($0.1625 vs $2.716 per million). Updated October 21, 2026; 30 benchmarks have a result for both.
DeepSeek V4 Pro 0423 or Gemma 3 12B: which to pick
Only DeepSeek V4 Pro 0423 has a coding score, 38.7 on the Vibe Coding Index; Gemma 3 12B has none, so the two cannot be ranked on coding.
Pick Gemma 3 12B for output tokens 16.7x cheaper ($0.1625 vs $2.716 per million).
- Shared results
- 30
- DeepSeek V4 Pro 0423 only
- 54
- Gemma 3 12B only
- 13
- 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
No call
Only DeepSeek V4 Pro 0423 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
DeepSeek V4 Pro 0423
DeepSeek V4 Pro 0423 leads, 30.4 against 3.8.
Same board, both measured
Agentic work
AA Agentic Index
DeepSeek V4 Pro 0423
DeepSeek V4 Pro 0423 leads, 26.3 against 0.1.
Same board, both measured
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
No call
Only DeepSeek V4 Pro 0423 has a Design Arena: websites result, 1234.
Directional only
Long documents
Listed context window
DeepSeek V4 Pro 0423
DeepSeek V4 Pro 0423 takes 1.05M tokens against 131K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Gemma 3 12B
DeepSeek V4 Pro 0423 $0.0026, Gemma 3 12B $0.0002 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 V4 Pro 0423 38.7 from 5 benchmarks, Gemma 3 12B not benchmarked.
- Widest category
- Math, 40.8 points apart on average over 3 shared benchmarks.
- Widest single gap
- OTIS Mock AIME: DeepSeek V4 Pro 0423 96.7% against Gemma 3 12B 16.7%, 80.0 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V4 Pro 0423 $2.716 against Gemma 3 12B $0.1625 per million output tokens, 16.7x apart.
- Context window
- DeepSeek V4 Pro 0423 takes 1.05M tokens, Gemma 3 12B 131K.
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 V4 Pro 0423 | Bars | Gemma 3 12B | Ties |
|---|---|---|---|---|
| Coding | 3 | 0 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 4 | 0 | ||
| Math | 3 | 0 | ||
| Knowledge | 4 | 0 | ||
| General | 13 | 0 | ||
| All shared boards | 30 | 0 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index30
| Benchmark | DeepSeek V4 Pro 0423 | Gemma 3 12B | Leader |
|---|---|---|---|
| LMArena WebDev | 1447 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| AA Coding Index | 59.4 | 5.8 | DeepSeek V4 Pro 0423 |
| LiveCodeBench (Vals AI) | 87.5% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| SWE-bench Verified (Vals AI) | 77.4% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Terminal-Bench 4 (Vals AI) | 11.1% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Vibe Code Bench (Vals AI) | 49.9% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Code migration (Vals AI) | 26.2% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| ProgramBench (Vals AI) | 0.0% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Terminal-Bench 2.1 (Vals AI) | 50.2% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| SkillsBench (Vals AI) | 51.3% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 13.7% | Only Gemma 3 12B measured |
| SciCode | Not measured | 16.4% | Only Gemma 3 12B measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 0.8% | Only Gemma 3 12B measured |
| FrontierCode | 17.6% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| WeirdML | 48.9% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| ALE-Bench | 1006.1 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: coding | 1470 | 1281 | DeepSeek V4 Pro 0423 |
| LMArena Text: software and IT | 1469 | 1313 | DeepSeek V4 Pro 0423 |
| LMArena WebDev: React | 1441 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena WebDev: HTML | 1435 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| SciCode | Not measured | 0.0% | Only Gemma 3 12B measured |
| Neither model has a result on the other 18 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | DeepSeek V4 Pro 0423 | Gemma 3 12B | Leader |
|---|---|---|---|
| AA Intelligence Index | 30.4 | 3.8 | DeepSeek V4 Pro 0423 |
| Epoch Capabilities Index | 149.1 | 123.5 | DeepSeek V4 Pro 0423 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | DeepSeek V4 Pro 0423 | Gemma 3 12B | Leader |
|---|---|---|---|
| AA Agentic Index | 26.3 | 0.1 | DeepSeek V4 Pro 0423 |
| Design Arena: agents building web apps | 1000 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: agents building full-stack apps | 948 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: agents building Godot games | 1059 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Finance Agent (Vals AI) | 44.1% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Tax Agent Bench (Vals AI) | 25.2% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Legal Research Bench (Vals AI) | 23.1% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Excel Modeling Benchmark (Vals AI) | 51.6% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| τ²-Bench Telecom | Not measured | 10.8% | Only Gemma 3 12B measured |
| Vending-Bench 2 | $3,285 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Berkeley Function Calling Leaderboard (BFCL) | Not measured | 30.4% | Only Gemma 3 12B measured |
| Neither model has a result on the other 30 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | DeepSeek V4 Pro 0423 | Gemma 3 12B | Leader |
|---|---|---|---|
| GPQA Diamond | 89.4% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| GPQA Diamond (Artificial Analysis run) | Not measured | 34.9% | Only Gemma 3 12B measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 4.2% | Only Gemma 3 12B measured |
| GPQA Diamond (Epoch AI run) | 90.9% | 39.5% | DeepSeek V4 Pro 0423 |
| Chess puzzles | 19.0% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Mystery game puzzles | 17.0% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: hard prompts | 1461 | 1309 | DeepSeek V4 Pro 0423 |
| LMArena Text: expert | 1463 | 1248 | DeepSeek V4 Pro 0423 |
| LMArena Text: hard prompts in English | 1466 | 1318 | DeepSeek V4 Pro 0423 |
| Kagi LLM Benchmark | 53.5% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Neither model has a result on the other 12 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index30
| Benchmark | DeepSeek V4 Pro 0423 | Gemma 3 12B | Leader |
|---|---|---|---|
| AA Math Index | Not measured | 18.3 | Only Gemma 3 12B measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 18.3% | Only Gemma 3 12B measured |
| OTIS Mock AIME | 96.7% | 16.7% | DeepSeek V4 Pro 0423 |
| FrontierMath, tiers 1 to 3 | 45.3% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| FrontierMath, tier 4 | 2.4% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| ProofBench | 16.0% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: math | 1443 | 1307 | DeepSeek V4 Pro 0423 |
| LMArena Text: mathematical work | 1455 | 1336 | 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 index40
| Benchmark | DeepSeek V4 Pro 0423 | Gemma 3 12B | Leader |
|---|---|---|---|
| MMLU Pro | 87.3% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LegalBench (Vals AI) | 80.3% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| CaseLaw (Vals AI) | 59.4% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| CorpFin (Vals AI) | 61.4% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| TaxEval (Vals AI) | 72.1% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| MedCode (Vals AI) | 40.5% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| MMLU Pro (Artificial Analysis run) | Not measured | 59.5% | Only Gemma 3 12B measured |
| SimpleQA Verified | 47.0% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: science | 1470 | 1366 | DeepSeek V4 Pro 0423 |
| LMArena Text: medicine and healthcare | 1466 | 1334 | DeepSeek V4 Pro 0423 |
| LMArena Text: legal and government | 1464 | 1381 | DeepSeek V4 Pro 0423 |
| LMArena Text: business and finance | 1443 | 1376 | DeepSeek V4 Pro 0423 |
| Neither model has a result on the other 6 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | DeepSeek V4 Pro 0423 | Gemma 3 12B | Leader |
|---|---|---|---|
| Design Arena: websites | 1234 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: UI components | 1222 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: game dev | 1229 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: data viz | 1203 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: 3D | 1254 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: all code categories | 1240 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: SVG | 1140 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| Design Arena: ASCII art | 1149 | Not measured | 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 index130
| Benchmark | DeepSeek V4 Pro 0423 | Gemma 3 12B | Leader |
|---|---|---|---|
| Public Benefits Bench (Vals AI) | 62.9% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| MedScribe (Vals AI) | 75.1% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| IFBench | Not measured | 36.7% | Only Gemma 3 12B measured |
| Long context reasoning (AA-LCR) | Not measured | 8.3% | Only Gemma 3 12B measured |
| DTBench | 90.7% | 48.8% | DeepSeek V4 Pro 0423 |
| LMCA | 41.2% | 4.5% | DeepSeek V4 Pro 0423 |
| Surface Evolver Bench | 40.0% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| CL-bench Life | 13.5% | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text | 1451 | 1334 | DeepSeek V4 Pro 0423 |
| LMArena Text: instruction following | 1445 | 1299 | DeepSeek V4 Pro 0423 |
| LMArena Text: creative writing | 1446 | 1331 | DeepSeek V4 Pro 0423 |
| LMArena Text: English | 1459 | 1349 | DeepSeek V4 Pro 0423 |
| LMArena Text: non-English | 1439 | 1318 | DeepSeek V4 Pro 0423 |
| LMArena Text: Chinese | 1481 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: German | 1458 | 1370 | DeepSeek V4 Pro 0423 |
| LMArena Text: Japanese | 1446 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: Korean | 1409 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: French | 1473 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: Spanish | 1457 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: Russian | 1453 | 1335 | DeepSeek V4 Pro 0423 |
| LMArena Text: Polish | 1461 | Not measured | Only DeepSeek V4 Pro 0423 measured |
| LMArena Text: multi-turn | 1468 | 1334 | DeepSeek V4 Pro 0423 |
| LMArena Text: longer queries | 1458 | 1317 | DeepSeek V4 Pro 0423 |
| LMArena Text: writing and language | 1447 | 1307 | DeepSeek V4 Pro 0423 |
| LMArena Text: entertainment and media | 1431 | 1297 | DeepSeek V4 Pro 0423 |
| 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 V4 Pro 0423 | Gemma 3 12B | 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
Gemma 3 12B costs less
Repository review
50K input + 3K output tokens
Gemma 3 12B costs less
Agent loop
200K input + 10K output tokens
Gemma 3 12B 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 V4 Pro 0423 | Google Gemma 3 12B |
|---|---|---|
| Maker | DeepSeek | |
| Weights | Open weights | Open weights |
| Released | April 24, 2026 | March 13, 2025 |
| Context window | 1.05M | 131K |
| Input $/Mtok | $1.225 | $0.075 |
| Output $/Mtok | $2.716 | $0.1625 |
| Coding index | 38.7 | Not benchmarked |
| Independent measurements behind it | 5 | 0 |
| Cost per solved task | $0.57 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V4 Pro 0423 better than Gemma 3 12B for coding?
Only DeepSeek V4 Pro 0423 has a coding score, 38.7 on the Vibe Coding Index; Gemma 3 12B has none, so the two cannot be ranked on coding. Pick Gemma 3 12B for output tokens 16.7x cheaper ($0.1625 vs $2.716 per million).
Which is cheaper, DeepSeek V4 Pro 0423 or Gemma 3 12B?
DeepSeek V4 Pro 0423 costs $1.225 per million input tokens and $2.716 per million output tokens; Gemma 3 12B costs $0.075 and $0.1625. Cost per solved task is known for DeepSeek V4 Pro 0423 only.
Which has the bigger context window, DeepSeek V4 Pro 0423 or Gemma 3 12B?
DeepSeek V4 Pro 0423, with 1.05M tokens against 131K.
How many benchmarks is the DeepSeek V4 Pro 0423 vs Gemma 3 12B comparison based on?
30 benchmarks have a result for both models as of October 21, 2026. DeepSeek V4 Pro 0423's coding index rests on 5 independent benchmarks and Gemma 3 12B's on 0. Where only one model has a result, the page says so instead of naming a winner.