Skip to main content

Gemma 4 31B (Reasoning)

GoogleGemmaOpen WeightApache 2.0 · Commercial OK

Description

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.

Release Date
2026-04-02
Parameters
30.7B
Context Length
262K
Modalities
image, text, video

Capability Radar

36
general
39
coding
86
reasoning
58
scienceest.
90
agents
70
multimodal

Science uses a reasoning proxy when dedicated science benchmarks are unavailable.

Rankings

Domain#RankScoreSource
Agents & Tools1
86.0
LS
Code Ranking63
68.0
AA
General Ranking90
69.0
AA
Science51
71.0
AA

Benchmark Scores (LLM Stats)

Agents

t2-bench86.4%SR

Biology

GPQA84.3%SR

Finance

MMLU-Pro85.2%SR

General

MMMLU88.4%SR
LiveCodeBench v680.0%SR
MMMU-Pro76.9%SR
BIG-Bench Extra Hard74.4%SR
MRCR v266.4%SR

Healthcare

MedXpertQA61.3%SR

Math

AIME 202689.2%SR
MathVision85.6%SR
Humanity's Last Exam26.5%SR

AA Evaluation Indices

Intelligence Index
39.2
Coding Index
38.7
Gpqa
0.9
Ifbench
0.8
Lcr
0.6
Tau2
0.6
Scicode
0.4
Terminalbench Hard
0.4
Hle
0.2

LLM Stats Category Scores

Tool Calling
90
Agents
90
Finance
90
Legal
90
Biology
80
Chemistry
80
General
80
Language
80
Physics
80
Math
70
Multimodal
70
Reasoning
70
Vision
60
Healthcare
60
Long Context
60

Pricing

Input PriceFree
Output PriceFree
Blended Price (3:1)Free

Speed

Tokens/sec34.9 tokens/s
Time to First Token1.02s
Time to Answer58.33s

Available Providers

(LS internal units)
ProviderInput PriceOutput Price
Novita140K400K

External Sources