Google model product
Gemma 4 31B is Google DeepMind's flagship dense multimodal model with 31 billion parameters and a 256K context window. Ranks #3 among open models on Arena AI. Built from the same research as Gemini 3, it features Per-Layer Embeddings, Shared KV Cache, alternating sliding-window and global attention, and variable aspect ratio vision encoding. Achieves an estimated LMArena text score of 1452.
Updated Aug 10, 2026. Default version: Gemma 4 31B
Structured fields from the published default version.
This profile uses the latest version under this unique model that has a calculated LLMBoard score. Arena and price are excluded.
Published benchmark records for the scored version Gemma 4 31B.
| BIG-Bench Extra Hard | 0.7 | 1 | 11 | 100.0% | C | |
| MedXpertQA | 0.6 | 5 | 12 | 63.6% | C | |
| MRCR v2 (8-needle) | 0.7 | 5 | 21 | 80.0% | C | |
| t2-bench | 0.9 | 5 | 23 | 81.8% | C | |
| MathVision | 0.9 | 9 | 32 | 74.2% | C | |
| AIME 2026 | 0.9 | 11 | 17 | 37.5% | C | |
| MMLU-Pro | 0.9 | 15 | 129 | 89.1% | C | |
| MMMLU | 0.9 | 18 | 49 | 64.6% | C | |
| LiveCodeBench v6 | 0.8 | 21 | 53 | 61.5% | C | |
| MMMU-Pro | 0.8 | 24 | 65 | 64.1% | C | |
| Humanity's Last Exam | 0.3 | 47 | 92 | 49.5% | C | |
| GPQA | 0.8 | 59 | 233 | 75.0% | C |
Preference and agent-evaluation signals from published Arena datasets.
| vision | captioning | 18 | 1208.2 | 75 | N/A | |
| vision | creative writing vision | 23 | 1278.4 | 1,904 | N/A | |
| vision | homework | 24 | 1306.2 | 5,412 | N/A | |
| vision style control | homework | 24 | 1300.4 | 5,412 | N/A | |
| vision style control | captioning | 25 | 1165.5 | 75 | N/A | |
| text style control | spanish | 26 | 1466.1 | 222 | N/A | |
| vision | entity recognition | 26 | 1219.1 | 288 | N/A | |
| vision | chinese | 27 | 1331.5 | 2,050 | N/A | |
| vision style control | chinese | 27 | 1311.5 | 2,050 | N/A | |
| vision style control | creative writing vision | 27 | 1259.1 | 1,904 | N/A | |
| vision style control | entity recognition | 27 | 1190.8 | 288 | N/A | |
| agent praise complaint | overall | 28 | -0.0 | N/A | 10.6K | |
| document style control | overall | 28 | 1445.2 | 10,577 | N/A | |
| text | russian | 28 | 1461.2 | 644 | N/A | |
| agent task outcome explicit | overall | 29 | -0.0 | N/A | 34.8K | |
| document | overall | 29 | 1423.7 | 10,577 | N/A | |
| vision | humor | 29 | 1270.9 | 1,253 | N/A | |
| vision style control | humor | 30 | 1250.8 | 1,253 | N/A | |
| vision | overall | 31 | 1270.1 | 32,461 | N/A | |
| text | spanish | 32 | 1457.2 | 222 | N/A | |
| vision | ocr | 32 | 1277.6 | 22,914 | N/A | |
| vision | diagram | 33 | 1281.2 | 8,375 | N/A | |
| text | math | 34 | 1466.0 | 398 | N/A | |
| vision | english | 34 | 1268.5 | 13,042 | N/A | |
| vision style control | english | 34 | 1254.0 | 13,042 | N/A | |
| vision style control | overall | 35 | 1255.7 | 32,461 | N/A | |
| vision style control | ocr | 35 | 1269.5 | 22,914 | N/A | |
| vision style control | diagram | 36 | 1281.4 | 8,375 | N/A | |
| text style control | math | 37 | 1471.3 | 398 | N/A | |
| text style control | russian | 38 | 1468.0 | 644 | N/A | |
| text style control | industry mathematical | 39 | 1474.1 | 319 | N/A | |
| agent steerability | overall | 40 | -0.1 | N/A | 42.8K | |
| text | industry mathematical | 40 | 1465.6 | 319 | N/A | |
| text factuality | hard prompts english | 44 | 1484.9 | 1,522 | N/A | |
| text style control | instruction following | 45 | 1452.1 | 1,660 | N/A | |
| agent | overall | 46 | -0.2 | N/A | 707.6K | |
| agent bash recovery steps | overall | 46 | -0.4 | N/A | 27.1K | |
| agent tool hallucination | overall | 46 | -0.3 | N/A | 592.3K | |
| text factuality | industry software and it services | 47 | 1487.1 | 2,099 | N/A | |
| text style control | longer query | 48 | 1466.8 | 1,643 | N/A | |
| text factuality | longer query | 49 | 1463.9 | 1,643 | N/A | |
| text style control | expert | 50 | 1482.8 | 446 | N/A | |
| text | industry writing and literature and language | 51 | 1429.4 | 1,338 | N/A | |
| text | multi turn | 51 | 1451.0 | 1,062 | N/A | |
| text factuality | instruction following | 51 | 1445.5 | 1,659 | N/A | |
| text factuality | exclude ties | 52 | 1450.8 | 4,005 | N/A | |
| text factuality | non english | 52 | 1437.6 | 3,262 | N/A | |
| text style control | hard prompts english | 52 | 1482.7 | 1,523 | N/A | |
| text | exclude ties | 53 | 1439.6 | 4,008 | N/A | |
| text | instruction following | 53 | 1431.7 | 1,660 | N/A | |
| text | non english | 53 | 1431.5 | 3,263 | N/A | |
| text factuality | overall | 53 | 1449.0 | 5,836 | N/A | |
| text style control | industry software and it services | 53 | 1490.5 | 2,100 | N/A | |
| text | longer query | 54 | 1443.1 | 1,643 | N/A | |
| text factuality | industry writing and literature and language | 54 | 1433.0 | 1,337 | N/A | |
| text | overall | 55 | 1441.5 | 5,839 | N/A | |
| text | creative writing | 55 | 1416.4 | 926 | N/A | |
| text | industry life and physical and social science | 55 | 1455.1 | 877 | N/A | |
| text factuality | coding | 55 | 1497.9 | 1,353 | N/A | |
| text factuality | hard prompts | 55 | 1470.8 | 3,351 | N/A | |
| text style control | industry life and physical and social science | 55 | 1473.3 | 877 | N/A | |
| text factuality | english | 56 | 1458.4 | 2,574 | N/A | |
| text | hard prompts english | 58 | 1453.9 | 1,523 | N/A | |
| text style control | overall | 58 | 1450.9 | 5,839 | N/A | |
| text style control | exclude ties | 58 | 1453.8 | 4,008 | N/A | |
| text style control | multi turn | 58 | 1464.2 | 1,062 | N/A | |
| text | industry software and it services | 59 | 1461.1 | 2,100 | N/A | |
| text style control | industry writing and literature and language | 59 | 1433.2 | 1,338 | N/A | |
| text | english | 60 | 1448.1 | 2,576 | N/A | |
| text | expert | 60 | 1451.6 | 446 | N/A | |
| text style control | hard prompts | 60 | 1473.2 | 3,352 | N/A | |
| text style control | non english | 60 | 1438.8 | 3,263 | N/A | |
| text factuality | creative writing | 61 | 1418.1 | 925 | N/A | |
| text style control | coding | 61 | 1498.1 | 1,354 | N/A | |
| text style control | english | 63 | 1459.8 | 2,576 | N/A | |
| text | hard prompts | 64 | 1445.5 | 3,352 | N/A | |
| text style control | industry legal and government | 64 | 1458.5 | 431 | N/A | |
| text | industry legal and government | 65 | 1442.4 | 431 | N/A | |
| text style control | industry entertainment and sports and media | 66 | 1416.0 | 1,031 | N/A | |
| text style control | creative writing | 67 | 1420.6 | 926 | N/A | |
| webdev | webdev-react | 69 | 1356.3 | 6,470 | N/A | |
| text | industry business and management and financial operations | 70 | 1426.8 | 1,129 | N/A | |
| text | industry entertainment and sports and media | 70 | 1403.8 | 1,031 | N/A | |
| text | industry medicine and healthcare | 71 | 1446.0 | 401 | N/A | |
| text style control | french | 71 | 1461.5 | 239 | N/A | |
| text style control | industry business and management and financial operations | 71 | 1444.6 | 1,129 | N/A | |
| text style control | industry medicine and healthcare | 71 | 1465.3 | 401 | N/A | |
| text | coding | 72 | 1455.2 | 1,354 | N/A | |
| webdev | webdev-html | 72 | 1373.9 | 911 | N/A | |
| text | chinese | 74 | 1470.4 | 372 | N/A | |
| webdev | overall | 74 | 1364.0 | 7,772 | N/A | |
| webdev | webdev | 74 | 1364.0 | 7,772 | N/A | |
| text style control | chinese | 83 | 1476.6 | 372 | N/A | |
| text | french | 84 | 1437.7 | 239 | N/A |
Official vendor API PAYG pricing is summarized first. The table then lists individual provider offerings without treating their minimum as the official price.
| Poe | google/gemma-4-31b | global | N/A | N/A | 262.1K | |
| Kenari | gemma-4-31b-it | global | N/A | N/A | 262.1K | |
| CrofAI | gemma-4-31b-it | global | $0.10 | $0.30 | 262.1K | |
| LLM Gateway | gemma-4-31b-it | global | $0.102 | $0.297 | 262.1K | |
| Neuralwatt | gemma-4-31b | global | $0.144 | $0.42 | 262.1K | |
| DigitalOcean | gemma-4-31B-it | global | $0.18 | $0.50 | 256K | |
| Cortecs | gemma-4-31b-it | global | $0.223 | $0.39 | 262K | |
| NanoGPT | TEE/gemma4-31b | global | $0.45 | $1 | 262.1K |
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
All published versions linked to this unique model. The score columns identify the version used by the current overall ranking.
| Gemma 4 31B | 74.3 | 30.7B | 262.1K | 32.8K | Yes | Apache 2.0 |
A concise description based on the published model registry.
Gemma 4 31B is Google DeepMind's flagship dense multimodal model with 31 billion parameters and a 256K context window. Ranks #3 among open models on Arena AI. Built from the same research as Gemini 3, it features Per-Layer Embeddings, Shared KV Cache, alternating sliding-window and global attention, and variable aspect ratio vision encoding. Achieves an estimated LMArena text score of 1452.
Use the benchmark, Arena and pricing sections above as separate evidence. A missing field means the current data snapshot does not support that claim.
Data snapshot: 2026-08-07. Editorial model content is not available in the backend.
Open a comparison with the three ranked models immediately above and below this model.
Recommendations prioritize the same model type and family, then the closest published LLMBoard score.
Common questions about Gemma 4 31B.
Gemma 4 31B's default version was released on Apr 2, 2026.
No official standard PAYG price is currently available for Gemma 4 31B. The lowest tracked third-party offer starts at $0.10 input and $0.30 output via CrofAI.
Gemma 4 31B is published under Google in the model registry.
The default version has a 262.1K token context window.
Yes. The default version is marked as open weight under Apache 2.0.
9 published provider offerings are linked to the default version.
Nearby ranked alternatives include DeepSeek-V3.2, Qwen3.5 397B A17B, Nova 2 Pro.