HolyGrAIL

All tasks / Models

Gemma 4 (31B)

Google DeepMind's open-weights Gemma line under Apache 2.0; Gemma 4 spans small on-device sizes up to a dense 31B.

  • Google DeepMind
  • Open weights
  • Released 2 Apr 2026

Official page

Where it ranks

  • A
    Chat & assistants
    Models · Local · 1,452 on LMArena Text Arena (overall, style control)
  • B
    Writing
    Models · Local · 1,419 on LMArena Text Arena: Creative Writing (style control)
  • –
    Documents & OCR
    Models · Local · no result on this task’s benchmarks yet
  • C
    Coding
    Models · Local · 1,368 on LMArena WebDev / Code Arena (overall)
  • A
    Image understanding
    Models · Local · 1,262 on LMArena Vision Arena (overall, style control)

Facts

Vendor
Google DeepMind
Family
Gemma
Released
2 Apr 2026
Weights
Open: download and run it yourself
License
Apache-2.0
Parameters
30.7B
GPU memory to run locally
~19 GB (31B dense: 4-bit ≈ 30.7 B × 0.57 GB + 1.5 GB ≈ 19 GB, estimate; Google publishes QAT q4_0 checkpoints. 26B-A4B MoE sibling also exists.)
API price
$0.14 in / $0.4 out per 1M tokens
Access
Free (open weights)
Last verified
18 Oct 2026

Price from openrouter.ai, 18 Oct 2026.

Benchmark results

BenchmarkResultEntry on the leaderboardAs ofSource
Artificial Analysis Intelligence Index15Gemma 4 31B (Reasoning)
Raw value from the page's embedded model data; the site displays it rounded to an integer.
18 Oct 2026artificialanalysis.ai
Artificial Analysis MMMU-Pro (independent evaluation)73.4%Gemma 4 31B (Reasoning)
Raw fraction 0.7341 from the model data embedded in https://artificialanalysis.ai/leaderboards/models, converted to percent; rank 58 of 124 entries with an MMMU-Pro result; no last-updated date shown.
19 Oct 2026artificialanalysis.ai
EQ-Bench Creative Writing v31,368google/gemma-4-31B-it
Elo score from the leaderboard data embedded in https://eqbench.com/creative_writing.js (file last modified 2026-09-24); rank 85 of 140; rubric score 16.01.
24 Sep 2026eqbench.com
LMArena Text Arena (overall, style control)1,452gemma-4-31b
From LMArena's official leaderboard-dataset (text_style_control, split latest, category overall, published 2026-10-15), which mirrors https://arena.ai/leaderboard/text (site blocks automated fetches).
15 Oct 2026huggingface.co
LMArena Text Arena: Creative Writing (style control)1,419gemma-4-31b
From LMArena's official leaderboard-dataset (text_style_control, split latest, category creative_writing, published 2026-10-15), which mirrors https://arena.ai/leaderboard/text/creative-writing (site blocks automated fetches). Rank 86 of 414; 974 votes; 95% CI 1400.0-1437.6.
15 Oct 2026huggingface.co
LMArena Vision Arena (overall, style control)1,262gemma-4-31b
From LMArena's official leaderboard-dataset (vision_style_control, split latest, category overall, published 2026-10-09), which mirrors https://arena.ai/leaderboard/vision (site blocks automated fetches). Rank 44 of 165; 39669 votes; 95% CI 1256.3-1267.9.
9 Oct 2026huggingface.co
LMArena WebDev / Code Arena (overall)1,368Gemma 4 31B
From LMArena's official leaderboard-dataset (webdev = Code Arena, split latest, category overall, published 2026-10-18), which mirrors https://arena.ai/leaderboard/webdev. Rank 107; 14592 votes; 95% CI 1361.7-1374.2.
18 Oct 2026huggingface.co
ParseBench (models)62.4%Gemma-4-31B-it
Overall (mean of tables, charts, content faithfulness, semantic formatting, visual grounding) from leaderboard.csv in the ParseBench repo (last leaderboard commit 2026-10-09).
9 Oct 2026github.com

Where you can use it

Earlier versions

  • Gemma 3 · 12 Mar 2025

Compare

Alternatives to Gemma 4 (31B)

Sources