DeepSeek V3.2 Exp vs Gemma 4 26B A4B
Gemma 4 26B A4B leads DeepSeek V3.2 Exp on the Artificial Analysis Intelligence Index, 16.7 to 16.6, and DeepSeek V3.2 Exp costs 1.5x less per output token ($0.41 vs $0.60 per million). Updated October 22, 2026; 37 benchmarks have a result for both.
DeepSeek V3.2 Exp or Gemma 4 26B A4B: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick DeepSeek V3.2 Exp for output tokens 1.5x cheaper ($0.41 vs $0.60 per million) and work measured by LMArena WebDev or LMArena Text: coding, where it leads.
- Shared results
- 37
- DeepSeek V3.2 Exp only
- 22
- Gemma 4 26B A4B only
- 18
- 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
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
Gemma 4 26B A4B
Gemma 4 26B A4B leads, 16.7 against 16.6.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Gemma 4 26B A4B has a AA Agentic Index result, 2.9.
Directional only
Web design
Design Arena: websites
No call
Only DeepSeek V3.2 Exp has a Design Arena: websites result, 1178.
Directional only
Long documents
Listed context window
Gemma 4 26B A4B
Gemma 4 26B A4B takes 262K tokens against 164K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Gemma 4 26B A4B
DeepSeek V3.2 Exp $0.0005, Gemma 4 26B A4B $0.0004 per turn.
List prices
At a glance
- Coding index
- DeepSeek V3.2 Exp not benchmarked, Gemma 4 26B A4B not benchmarked.
- Widest category
- Agentic work, 9.6 points apart on average over 1 shared benchmark.
- Widest single gap
- IFBench: DeepSeek V3.2 Exp 54.1% against Gemma 4 26B A4B 72.4%, 18.3 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V3.2 Exp $0.41 against Gemma 4 26B A4B $0.60 per million output tokens, 1.5x apart.
- Context window
- DeepSeek V3.2 Exp takes 164K tokens, Gemma 4 26B A4B 262K.
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 | Gemma 4 26B A4B | Ties |
|---|---|---|---|---|
| Coding | 2 | 4 | ||
| Intelligence | 1 | 1 | ||
| Agentic work | 0 | 1 | ||
| Reasoning | 1 | 4 | ||
| Math | 0 | 2 | ||
| Knowledge | 1 | 3 | ||
| General | 6 | 11 | ||
| All shared boards | 11 | 26 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index24
| Benchmark | DeepSeek V3.2 Exp | Gemma 4 26B A4B | Leader |
|---|---|---|---|
| Aider polyglot | 74.2% | Not measured | Only DeepSeek V3.2 Exp measured |
| LMArena WebDev | 1272 | 1360 | Gemma 4 26B A4B |
| AA Coding Index | Not measured | 39.3 | Only Gemma 4 26B A4B measured |
| LiveCodeBench (Artificial Analysis run) | 78.9% | Not measured | Only DeepSeek V3.2 Exp measured |
| SciCode | Not measured | 40.0% | Only Gemma 4 26B A4B measured |
| Terminal-Bench Hard (Artificial Analysis run) | 31.1% | 25.0% | DeepSeek V3.2 Exp |
| WeirdML | 39.5% | 35.2% | DeepSeek V3.2 Exp |
| ALE-Bench | Not measured | 927.2 | Only Gemma 4 26B A4B measured |
| LMArena Text: coding | 1438 | 1446 | Gemma 4 26B A4B |
| LMArena Text: software and IT | 1438 | 1454 | Gemma 4 26B A4B |
| LMArena WebDev: React | Not measured | 1351 | Only Gemma 4 26B A4B measured |
| LMArena WebDev: HTML | 1281 | 1362 | Gemma 4 26B A4B |
| Neither model has a result on the other 27 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index11
| Benchmark | DeepSeek V3.2 Exp | Gemma 4 26B A4B | Leader |
|---|---|---|---|
| AA Intelligence Index | 16.6 | 16.7 | Gemma 4 26B A4B |
| Epoch Capabilities Index | 144.9 | 141.9 | DeepSeek V3.2 Exp |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | DeepSeek V3.2 Exp | Gemma 4 26B A4B | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 2.9 | Only Gemma 4 26B A4B measured |
| τ²-Bench Telecom | 33.9% | 43.6% | Gemma 4 26B A4B |
| TheAgentCompany | 42.9% | Not measured | Only DeepSeek V3.2 Exp measured |
| Berkeley Function Calling Leaderboard (BFCL) | 56.7% | Not measured | Only DeepSeek V3.2 Exp measured |
| Neither model has a result on the other 37 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index14
| Benchmark | DeepSeek V3.2 Exp | Gemma 4 26B A4B | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 79.7% | 79.2% | DeepSeek V3.2 Exp |
| Humanity's Last Exam (Artificial Analysis run) | 14.9% | 19.3% | Gemma 4 26B A4B |
| GPQA Diamond (Epoch AI run) | Not measured | 73.2% | Only Gemma 4 26B A4B measured |
| LMArena Text: hard prompts | 1429 | 1438 | Gemma 4 26B A4B |
| LMArena Text: expert | 1422 | 1447 | Gemma 4 26B A4B |
| LMArena Text: hard prompts in English | 1444 | 1448 | Gemma 4 26B A4B |
| LMArena Vision: diagrams | Not measured | 1264 | Only Gemma 4 26B A4B measured |
| LMArena Vision: homework | Not measured | 1293 | Only Gemma 4 26B A4B measured |
| Kagi LLM Benchmark | 52.2% | Not measured | Only DeepSeek V3.2 Exp measured |
| Neither model has a result on the other 13 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V3.2 Exp | Gemma 4 26B A4B | 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 | 82.2% | Only Gemma 4 26B A4B measured |
| LMArena Text: math | 1422 | 1470 | Gemma 4 26B A4B |
| LMArena Text: mathematical work | 1433 | 1475 | Gemma 4 26B A4B |
| Neither model has a result on the other 8 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index13
| Benchmark | DeepSeek V3.2 Exp | Gemma 4 26B A4B | Leader |
|---|---|---|---|
| MMLU Pro (Artificial Analysis run) | 85.0% | Not measured | Only DeepSeek V3.2 Exp measured |
| LMArena Text: science | 1443 | 1446 | Gemma 4 26B A4B |
| LMArena Text: medicine and healthcare | 1468 | 1428 | DeepSeek V3.2 Exp |
| LMArena Text: legal and government | 1419 | 1457 | Gemma 4 26B A4B |
| LMArena Text: business and finance | 1412 | 1429 | Gemma 4 26B A4B |
| FACTS Benchmark Suite | Not measured | 42.7% | Only Gemma 4 26B A4B measured |
| Neither model has a result on the other 12 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index7 boards, none shared
| Benchmark | DeepSeek V3.2 Exp | Gemma 4 26B A4B | Leader |
|---|---|---|---|
| Design Arena: websites | 1178 | Not measured | Only DeepSeek V3.2 Exp measured |
| Design Arena: UI components | 1164 | Not measured | Only DeepSeek V3.2 Exp measured |
| Design Arena: game dev | 1144 | Not measured | Only DeepSeek V3.2 Exp measured |
| Design Arena: data viz | 1152 | Not measured | Only DeepSeek V3.2 Exp measured |
| Design Arena: 3D | 1148 | Not measured | Only DeepSeek V3.2 Exp measured |
| Design Arena: all code categories | 1170 | Not measured | Only DeepSeek V3.2 Exp measured |
| Design Arena: SVG | 1031 | Not measured | Only DeepSeek V3.2 Exp measured |
| Neither model has a result on the other 2 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index611
| Benchmark | DeepSeek V3.2 Exp | Gemma 4 26B A4B | Leader |
|---|---|---|---|
| IFBench | 54.1% | 72.4% | Gemma 4 26B A4B |
| Long context reasoning (AA-LCR) | 72.3% | 65.7% | DeepSeek V3.2 Exp |
| DTBench | 87.7% | 74.9% | DeepSeek V3.2 Exp |
| LMCA | 29.1% | 29.7% | Gemma 4 26B A4B |
| Fiction.LiveBench | 83.3% | Not measured | Only DeepSeek V3.2 Exp measured |
| CL-bench | 13.2% | Not measured | Only DeepSeek V3.2 Exp measured |
| CL-bench Life | 9.5% | Not measured | Only DeepSeek V3.2 Exp measured |
| LMArena Text | 1425 | 1434 | Gemma 4 26B A4B |
| LMArena Text: instruction following | 1401 | 1420 | Gemma 4 26B A4B |
| LMArena Text: creative writing | 1404 | 1400 | DeepSeek V3.2 Exp |
| LMArena Vision | Not measured | 1260 | Only Gemma 4 26B A4B measured |
| LMArena Text: English | 1441 | 1444 | Gemma 4 26B A4B |
| LMArena Text: non-English | 1404 | 1421 | Gemma 4 26B A4B |
| LMArena Text: Chinese | 1461 | 1495 | Gemma 4 26B A4B |
| LMArena Text: German | 1440 | Not measured | Only DeepSeek V3.2 Exp measured |
| LMArena Text: Japanese | 1331 | Not measured | Only DeepSeek V3.2 Exp measured |
| LMArena Text: Korean | 1372 | Not measured | Only DeepSeek V3.2 Exp measured |
| LMArena Text: French | 1434 | 1460 | Gemma 4 26B A4B |
| LMArena Text: Spanish | 1441 | 1415 | DeepSeek V3.2 Exp |
| LMArena Text: Russian | 1412 | 1434 | Gemma 4 26B A4B |
| LMArena Text: Polish | 1435 | Not measured | Only DeepSeek V3.2 Exp measured |
| LMArena Text: multi-turn | 1422 | 1441 | Gemma 4 26B A4B |
| LMArena Text: longer queries | 1418 | 1427 | Gemma 4 26B A4B |
| LMArena Text: writing and language | 1407 | 1405 | DeepSeek V3.2 Exp |
| LMArena Text: entertainment and media | 1407 | 1403 | DeepSeek V3.2 Exp |
| LMArena Vision: English | Not measured | 1256 | Only Gemma 4 26B A4B measured |
| LMArena Vision: Chinese | Not measured | 1312 | Only Gemma 4 26B A4B measured |
| LMArena Vision: reading text | Not measured | 1268 | Only Gemma 4 26B A4B measured |
| LMArena Vision: humor | Not measured | 1272 | Only Gemma 4 26B A4B measured |
| LMArena Vision: captioning | Not measured | 1297 | Only Gemma 4 26B A4B measured |
| LMArena Vision: recognizing things | Not measured | 1191 | Only Gemma 4 26B A4B measured |
| LMArena Vision: creative writing | Not measured | 1256 | Only Gemma 4 26B A4B measured |
| Neither model has a result on the other 14 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V3.2 Exp | Gemma 4 26B A4B | 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 4 26B A4B costs less
Repository review
50K input + 3K output tokens
Gemma 4 26B A4B costs less
Agent loop
200K input + 10K output tokens
Gemma 4 26B A4B 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 | Google Gemma 4 26B A4B |
|---|---|---|
| Maker | DeepSeek | |
| Weights | Open weights | Open weights |
| Released | September 29, 2025 | April 3, 2026 |
| Context window | 164K | 262K |
| Input $/Mtok | $0.27 | $0.15 |
| Output $/Mtok | $0.41 | $0.60 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 1 | 1 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V3.2 Exp better than Gemma 4 26B A4B for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick DeepSeek V3.2 Exp for output tokens 1.5x cheaper ($0.41 vs $0.60 per million) and work measured by LMArena WebDev or LMArena Text: coding, where it leads.
Which is cheaper, DeepSeek V3.2 Exp or Gemma 4 26B A4B?
DeepSeek V3.2 Exp costs $0.27 per million input tokens and $0.41 per million output tokens; Gemma 4 26B A4B costs $0.15 and $0.60. Cost per solved task is known for neither of them.
Which has the bigger context window, DeepSeek V3.2 Exp or Gemma 4 26B A4B?
Gemma 4 26B A4B, with 262K tokens against 164K.
How many benchmarks is the DeepSeek V3.2 Exp vs Gemma 4 26B A4B comparison based on?
37 benchmarks have a result for both models as of October 22, 2026. DeepSeek V3.2 Exp's coding index rests on 1 independent benchmark and Gemma 4 26B A4B's on 1. Where only one model has a result, the page says so instead of naming a winner.