Passer au contenu principal

Gemma 4 31B (Reasoning)

GoogleGemmaOpen WeightApache 2.0 · Usage Commercial

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.

Date de sortie
2026-04-02
Paramètres
30.7B
Longueur du contexte
262K
Modalités
audio, image, text, video

Radar de capacités

17
general
44
coding
86
reasoning
59
science
90
agents
70
multimodal

Classements

Domaine#RangScoreSource
Capacité agentique23
58.0
LS
Classement codage239
60.0
AA
Classement général248
48.0
AA
Classement multimodal65
54.0
LS
Science172
63.0
AA

Scores de benchmarks (LLM Stats)

(LLM Stats (zeroeval))

Agents

t2-bench86.4%Aut.

Healthcare

MedXpertQA61.3%Aut.

Language

MMMLU88.4%Aut.
MMLU-Pro85.2%Aut.

Long Context

MRCR v2 (8-needle)66.4%Aut.

Math

AIME 202689.2%Aut.
MathVision85.6%Aut.

Reasoning

GPQANYU + Cohere + Anthropic (2023)84.3%Aut.
LiveCodeBench v680.0%Aut.
BIG-Bench Extra Hard74.4%Aut.
Humanity's Last Exam26.5%Aut.

Vision

MMMU-Pro76.9%Aut.

Indices d'évaluation AA

(Artificial Analysis)
Gpqa(NYU + Cohere + Anthropic (2023))
85.7
Ifbench(Google Research (2023))
75.6
Lcr(Artificial Analysis)
69.7
Tau2(Sierra + U Toronto + Vector Institute (2025))
59.9
Scicode(UIUC + Argonne National Lab (2024))
45.5
Terminalbench V2 1
43.4
Coding Index(Artificial Analysis)
43.4
Terminalbench Hard(Stanford × Laude Institute (2026))
36.4
Hle(Center for AI Safety + Scale AI (2025))
23.6
Tau Banking
14.8
Intelligence Index(Artificial Analysis)
14.7
Terminalbench V4 0
0.0

Scores par catégorie LLM Stats

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

Tarification

Prix d'entréeGratuit
Prix de sortieGratuit
Prix mixte (3:1)Gratuit

Vitesse

Tokens/sec35.0
Délai du premier token0.98s
Temps de réponse50.59s

Classement des Prix par Fournisseur

Classement des Prix par Fournisseur

21 fournisseurs

Moins cher: DeepInfraPlus cher: Opper
FournisseurEntréeSortie
1DeepInfraMoins cher
$0
$0
2Novita
$0
$0
3FriendliAI
$0
$0
4Together
$0
$0
5Kilo Gateway
$0.08
$0.3
6OpenRouter
$0.09
$0.34
7NanoGPT
$0.1
$0.45
8DevPass (LLM Gateway)
$0.1
$0.25
9CrofAI
$0.1
$0.3
10Lilac
$0.11
$0.35
11FastRouter
$0.13
$0.38
12OrcaRouter
$0.13
$0.38
13Abacus
$0.14
$0.4
14NovitaAI
$0.14
$0.4
15Vercel AI Gateway
$0.14
$0.4
16Merge Gateway
$0.14
$0.4
17Crusoe
$0.14
$0.4
18Neuralwatt
$0.144
$0.42
19ai&
$0.2
$0.5
20Cortecs
$0.223
$0.39
21Opper
$0.46488
$2.44062

Comparer les prix entre différents fournisseurs API pour ce modèle.

Sources externes