Gemma 4 31B vs Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads Gemma 4 31B on coding, 62.4 to 39.0 on the Vibe Coding Index, and Gemma 4 31B costs 15.0x less per output token ($0.40 vs $6.00 per million). Updated October 22, 2026; 58 benchmarks have a result for both.
Gemma 4 31B or Qwen3.8 Max (0902): which to pick
Pick Qwen3.8 Max (0902) for coding: it scores 62.4 on the Vibe Coding Index against 39.0 for Gemma 4 31B.
Pick Gemma 4 31B for output tokens 15.0x cheaper ($0.40 vs $6.00 per million) and work measured by SAGE (Vals AI), where it leads.
- Shared results
- 58
- Gemma 4 31B only
- 13
- Qwen3.8 Max (0902) only
- 55
- Like-for-like categories
- 8 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
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) scores 62.4 against 39.0.
2 and 5 benchmarks behind the scores
Intelligence
AA Intelligence Index
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 45.4 against 14.7.
Same board, both measured
Agentic work
AA Agentic Index
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 56.0 against 4.2.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only Qwen3.8 Max (0902) has a GPQA Diamond result, 93.7%.
Directional only
Knowledge
MMLU Pro
No call
Only Qwen3.8 Max (0902) has a MMLU Pro result, 88.6%.
Directional only
Web design
Design Arena: websites
No call
Only Qwen3.8 Max (0902) has a Design Arena: websites result, 1275.
Directional only
Long documents
Listed context window
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) takes 1M tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Gemma 4 31B
Gemma 4 31B $0.0003, Qwen3.8 Max (0902) $0.0050 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Qwen3.8 Max (0902) has the number, $1.32.
Directional only
At a glance
- Coding index
- Gemma 4 31B 39.0 from 2 benchmarks, Qwen3.8 Max (0902) 62.4 from 5 benchmarks.
- Widest category
- Agentic work, 33.6 points apart on average over 4 shared benchmarks.
- Widest single gap
- AA Agentic Index: Gemma 4 31B 4.2 against Qwen3.8 Max (0902) 56.0, 51.8 points apart on a 0 to 100 scale.
- Output price
- Gemma 4 31B $0.40 against Qwen3.8 Max (0902) $6.00 per million output tokens, 15.0x apart.
- Context window
- Gemma 4 31B takes 262K tokens, Qwen3.8 Max (0902) 1M.
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 | Gemma 4 31B | Bars | Qwen3.8 Max (0902) | Ties |
|---|---|---|---|---|
| Coding | 0 | 8 | ||
| Intelligence | 0 | 2 | ||
| Agentic work | 0 | 4 | ||
| Reasoning | 0 | 9 | ||
| Math | 0 | 3 | ||
| Knowledge | 0 | 6 | ||
| Web design | 0 | 1 | ||
| General | 1 | 24 | ||
| All shared boards | 1 | 57 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index08
| Benchmark | Gemma 4 31B | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| LiveBench coding | 49.0% | 60.3% | Qwen3.8 Max (0902) |
| LMArena WebDev | 1367 | 1674 | Qwen3.8 Max (0902) |
| AA Coding Index | 43.4 | 76.2 | Qwen3.8 Max (0902) |
| LiveCodeBench (Vals AI) | Not measured | 87.8% | Only Qwen3.8 Max (0902) measured |
| SWE-bench Verified (Vals AI) | Not measured | 85.6% | Only Qwen3.8 Max (0902) measured |
| Terminal-Bench 4 (Vals AI) | Not measured | 34.3% | Only Qwen3.8 Max (0902) measured |
| Vibe Code Bench (Vals AI) | Not measured | 64.7% | Only Qwen3.8 Max (0902) measured |
| Code migration (Vals AI) | Not measured | 24.0% | Only Qwen3.8 Max (0902) measured |
| ProgramBench (Vals AI) | Not measured | 0.0% | Only Qwen3.8 Max (0902) measured |
| IOI (Vals AI) | Not measured | 68.9% | Only Qwen3.8 Max (0902) measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 67.4% | Only Qwen3.8 Max (0902) measured |
| SWE-bench Multimodal (Vals AI) | Not measured | 42.5% | Only Qwen3.8 Max (0902) measured |
| Vibe Code Bench 1-100 (Vals AI) | Not measured | 12.8% | Only Qwen3.8 Max (0902) measured |
| SkillsBench (Vals AI) | Not measured | 42.0% | Only Qwen3.8 Max (0902) measured |
| CyberBench (Vals AI) | Not measured | 28.6% | Only Qwen3.8 Max (0902) measured |
| SciCode | 45.5% | 53.2% | Qwen3.8 Max (0902) |
| Terminal-Bench Hard (Artificial Analysis run) | 36.4% | Not measured | Only Gemma 4 31B measured |
| FrontierSWE | Not measured | 17.8% | Only Qwen3.8 Max (0902) measured |
| WeirdML | 52.3% | Not measured | Only Gemma 4 31B measured |
| ALE-Bench | 925.5 | Not measured | Only Gemma 4 31B measured |
| LMArena Text: coding | 1459 | 1502 | Qwen3.8 Max (0902) |
| LMArena Text: software and IT | 1463 | 1498 | Qwen3.8 Max (0902) |
| LMArena WebDev: React | 1359 | 1692 | Qwen3.8 Max (0902) |
| LMArena WebDev: HTML | 1380 | 1649 | Qwen3.8 Max (0902) |
| LMArena WebDev: from an image | Not measured | 1637 | Only Qwen3.8 Max (0902) measured |
| SciCode | 8.9% | Not measured | Only Gemma 4 31B measured |
| Neither model has a result on the other 13 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Gemma 4 31B | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| AA Intelligence Index | 14.7 | 45.4 | Qwen3.8 Max (0902) |
| Epoch Capabilities Index | 143.0 | 157.4 | Qwen3.8 Max (0902) |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | Gemma 4 31B | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| AA Agentic Index | 4.2 | 56.0 | Qwen3.8 Max (0902) |
| Design Arena: agents building web apps | Not measured | 1300 | Only Qwen3.8 Max (0902) measured |
| Design Arena: agents building full-stack apps | Not measured | 1305 | Only Qwen3.8 Max (0902) measured |
| Design Arena: agents building mobile apps | Not measured | 1224 | Only Qwen3.8 Max (0902) measured |
| Design Arena: agents building Android apps | Not measured | 1196 | Only Qwen3.8 Max (0902) measured |
| Design Arena: agents building Godot games | Not measured | 1234 | Only Qwen3.8 Max (0902) measured |
| Design Arena: PowerPoint slides | Not measured | 1233 | Only Qwen3.8 Max (0902) measured |
| Design Arena: spreadsheets | Not measured | 967 | Only Qwen3.8 Max (0902) measured |
| LiveBench research | 44.4% | 69.9% | Qwen3.8 Max (0902) |
| LiveBench browser | 4.0% | 32.0% | Qwen3.8 Max (0902) |
| LiveBench long agentic | 13.3% | 42.3% | Qwen3.8 Max (0902) |
| Finance Agent (Vals AI) | Not measured | 50.6% | Only Qwen3.8 Max (0902) measured |
| Tax Agent Bench (Vals AI) | Not measured | 32.1% | Only Qwen3.8 Max (0902) measured |
| Legal Research Bench (Vals AI) | Not measured | 47.6% | Only Qwen3.8 Max (0902) measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 60.1% | Only Qwen3.8 Max (0902) measured |
| Terminal-Bench-Science (Vals AI) | Not measured | 12.9% | Only Qwen3.8 Max (0902) measured |
| τ²-Bench Telecom | 65.5% | Not measured | Only Gemma 4 31B measured |
| APEX-Agents | Not measured | 63.3% | Only Qwen3.8 Max (0902) measured |
| LMArena Search | Not measured | 1261 | Only Qwen3.8 Max (0902) measured |
| Neither model has a result on the other 22 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index09
| Benchmark | Gemma 4 31B | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 93.7% | Only Qwen3.8 Max (0902) measured |
| LiveBench reasoning | 68.3% | 88.2% | Qwen3.8 Max (0902) |
| MysteryMechanism (Vals AI) | Not measured | 23.9% | Only Qwen3.8 Max (0902) measured |
| MMMU Pro (Vals AI) | Not measured | 88.0% | Only Qwen3.8 Max (0902) measured |
| GPQA Diamond (Artificial Analysis run) | 85.7% | 92.8% | Qwen3.8 Max (0902) |
| Humanity's Last Exam (Artificial Analysis run) | 23.6% | 43.1% | Qwen3.8 Max (0902) |
| GPQA Diamond (Epoch AI run) | 75.8% | 92.7% | Qwen3.8 Max (0902) |
| Chess puzzles | Not measured | 46.0% | Only Qwen3.8 Max (0902) measured |
| Mystery game puzzles | Not measured | 38.0% | Only Qwen3.8 Max (0902) measured |
| LMArena Text: hard prompts | 1448 | 1496 | Qwen3.8 Max (0902) |
| LMArena Text: expert | 1464 | 1507 | Qwen3.8 Max (0902) |
| LMArena Text: hard prompts in English | 1456 | 1503 | Qwen3.8 Max (0902) |
| LMArena Vision: diagrams | 1283 | 1323 | Qwen3.8 Max (0902) |
| LMArena Vision: homework | 1310 | 1336 | Qwen3.8 Max (0902) |
| Kagi LLM Benchmark | 63.5% | Not measured | Only Gemma 4 31B measured |
| Neither model has a result on the other 7 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Gemma 4 31B | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| OTIS Mock AIME | 73.3% | 100.0% | Qwen3.8 Max (0902) |
| FrontierMath, tiers 1 to 3 | Not measured | 74.7% | Only Qwen3.8 Max (0902) measured |
| FrontierMath, tier 4 | Not measured | 46.3% | Only Qwen3.8 Max (0902) measured |
| ProofBench | Not measured | 58.0% | Only Qwen3.8 Max (0902) measured |
| LMArena Text: math | 1463 | 1501 | Qwen3.8 Max (0902) |
| LMArena Text: mathematical work | 1465 | 1491 | Qwen3.8 Max (0902) |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index06
| Benchmark | Gemma 4 31B | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 88.6% | Only Qwen3.8 Max (0902) measured |
| LegalBench (Vals AI) | Not measured | 83.6% | Only Qwen3.8 Max (0902) measured |
| CaseLaw (Vals AI) | 52.6% | Not measured | Only Gemma 4 31B measured |
| CorpFin (Vals AI) | Not measured | 65.8% | Only Qwen3.8 Max (0902) measured |
| TaxEval (Vals AI) | Not measured | 75.5% | Only Qwen3.8 Max (0902) measured |
| MortgageTax (Vals AI) | 61.4% | 64.0% | Qwen3.8 Max (0902) |
| MedCode (Vals AI) | Not measured | 40.7% | Only Qwen3.8 Max (0902) measured |
| SimpleQA Verified | 10.4% | 47.3% | Qwen3.8 Max (0902) |
| LMArena Text: science | 1457 | 1512 | Qwen3.8 Max (0902) |
| LMArena Text: medicine and healthcare | 1451 | 1503 | Qwen3.8 Max (0902) |
| LMArena Text: legal and government | 1438 | 1492 | Qwen3.8 Max (0902) |
| LMArena Text: business and finance | 1430 | 1480 | Qwen3.8 Max (0902) |
| FACTS Benchmark Suite | 45.3% | Not measured | Only Gemma 4 31B measured |
| Neither model has a result on the other 5 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Gemma 4 31B | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1275 | Only Qwen3.8 Max (0902) measured |
| Design Arena: UI components | Not measured | 1338 | Only Qwen3.8 Max (0902) measured |
| Design Arena: game dev | Not measured | 1312 | Only Qwen3.8 Max (0902) measured |
| Design Arena: data viz | Not measured | 1281 | Only Qwen3.8 Max (0902) measured |
| Design Arena: 3D | Not measured | 1336 | Only Qwen3.8 Max (0902) measured |
| Design Arena: all code categories | Not measured | 1293 | Only Qwen3.8 Max (0902) measured |
| Design Arena: ASCII art | Not measured | 1200 | Only Qwen3.8 Max (0902) measured |
| LiveBench design | 58.4% | 78.6% | Qwen3.8 Max (0902) |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index124
| Benchmark | Gemma 4 31B | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| LiveBench data analysis | 53.3% | 78.4% | Qwen3.8 Max (0902) |
| Public Benefits Bench (Vals AI) | Not measured | 67.1% | Only Qwen3.8 Max (0902) measured |
| SAGE (Vals AI) | 55.0% | 51.2% | Gemma 4 31B |
| MedScribe (Vals AI) | Not measured | 85.0% | Only Qwen3.8 Max (0902) measured |
| Vals Multimodal Index | Not measured | 65.4% | Only Qwen3.8 Max (0902) measured |
| IFBench | 75.6% | Not measured | Only Gemma 4 31B measured |
| Long context reasoning (AA-LCR) | 69.7% | 80.3% | Qwen3.8 Max (0902) |
| DTBench | 82.7% | 92.0% | Qwen3.8 Max (0902) |
| LMCA | 39.3% | 46.2% | Qwen3.8 Max (0902) |
| Surface Evolver Bench | 30.6% | Not measured | Only Gemma 4 31B measured |
| Furniture assembly | Not measured | 20.0% | Only Qwen3.8 Max (0902) measured |
| LMArena Text | 1443 | 1483 | Qwen3.8 Max (0902) |
| LMArena Text: instruction following | 1433 | 1481 | Qwen3.8 Max (0902) |
| LMArena Text: creative writing | 1417 | 1477 | Qwen3.8 Max (0902) |
| LMArena Vision | 1276 | 1313 | Qwen3.8 Max (0902) |
| LMArena Text: English | 1451 | 1491 | Qwen3.8 Max (0902) |
| LMArena Text: non-English | 1430 | 1473 | Qwen3.8 Max (0902) |
| LMArena Text: Chinese | 1475 | 1539 | Qwen3.8 Max (0902) |
| LMArena Text: German | Not measured | 1478 | Only Qwen3.8 Max (0902) measured |
| LMArena Text: Japanese | Not measured | 1473 | Only Qwen3.8 Max (0902) measured |
| LMArena Text: Korean | Not measured | 1460 | Only Qwen3.8 Max (0902) measured |
| LMArena Text: French | 1436 | 1498 | Qwen3.8 Max (0902) |
| LMArena Text: Spanish | 1442 | 1489 | Qwen3.8 Max (0902) |
| LMArena Text: Russian | 1460 | 1482 | Qwen3.8 Max (0902) |
| LMArena Text: Polish | 1433 | 1471 | Qwen3.8 Max (0902) |
| LMArena Text: multi-turn | 1452 | 1489 | Qwen3.8 Max (0902) |
| LMArena Text: longer queries | 1446 | 1488 | Qwen3.8 Max (0902) |
| LMArena Text: writing and language | 1432 | 1477 | Qwen3.8 Max (0902) |
| LMArena Text: entertainment and media | 1405 | 1464 | Qwen3.8 Max (0902) |
| LMArena Vision: English | 1271 | 1311 | Qwen3.8 Max (0902) |
| LMArena Vision: Chinese | 1326 | 1358 | Qwen3.8 Max (0902) |
| LMArena Vision: reading text | 1282 | 1322 | Qwen3.8 Max (0902) |
| LMArena Vision: humor | 1280 | 1304 | Qwen3.8 Max (0902) |
| LMArena Vision: captioning | 1218 | Not measured | Only Gemma 4 31B measured |
| LMArena Vision: recognizing things | 1226 | Not measured | Only Gemma 4 31B measured |
| LMArena Vision: creative writing | 1285 | 1312 | Qwen3.8 Max (0902) |
| LMArena Document | 1425 | Not measured | Only Gemma 4 31B measured |
| Neither model has a result on the other 9 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Gemma 4 31B | Qwen3.8 Max (0902) | 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 31B costs less
Repository review
50K input + 3K output tokens
Gemma 4 31B costs less
Agent loop
200K input + 10K output tokens
Gemma 4 31B 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 | Google Gemma 4 31B | Qwen3.8 Max (0902) |
|---|---|---|
| Maker | Qwen | |
| Weights | Open weights | No public weights listed |
| Released | April 2, 2026 | September 3, 2026 |
| Context window | 262K | 1M |
| Input $/Mtok | $0.14 | $2.00 |
| Output $/Mtok | $0.40 | $6.00 |
| Coding index | 39.0 | 62.4 |
| Independent measurements behind it | 2 | 5 |
| Cost per solved task | Not computable | $1.32 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Gemma 4 31B better than Qwen3.8 Max (0902) for coding?
Pick Qwen3.8 Max (0902) for coding: it scores 62.4 on the Vibe Coding Index against 39.0 for Gemma 4 31B. Pick Gemma 4 31B for output tokens 15.0x cheaper ($0.40 vs $6.00 per million) and work measured by SAGE (Vals AI), where it leads.
Which is cheaper, Gemma 4 31B or Qwen3.8 Max (0902)?
Gemma 4 31B costs $0.14 per million input tokens and $0.40 per million output tokens; Qwen3.8 Max (0902) costs $2.00 and $6.00. Cost per solved task is known for Qwen3.8 Max (0902) only.
Which has the bigger context window, Gemma 4 31B or Qwen3.8 Max (0902)?
Qwen3.8 Max (0902), with 1M tokens against 262K.
How many benchmarks is the Gemma 4 31B vs Qwen3.8 Max (0902) comparison based on?
58 benchmarks have a result for both models as of October 22, 2026. Gemma 4 31B's coding index rests on 2 independent benchmarks and Qwen3.8 Max (0902)'s on 5. Where only one model has a result, the page says so instead of naming a winner.