Gemma 4 E2B

Fiche modèle · Google · gemma-4-e2b-it

Gemma 4 E2B is Google DeepMind's smallest multimodal model with 2.3 billion effective parameters (5.1B with embeddings) and a 128K context window. Supports image, text, and audio inputs. Designed for on-device and edge deployment with Per-Layer Embeddings for efficient inference.

Identité

OrganisationGoogle
Pays—
Famille—
Typellm
Date de sortie2026-04-02
Licenceapache 2.0
Licence commercialeoui
Licence open sourceoui
Poids ouvertsoui
Auto-hébergement (DGX Spark)oui — ≤ 107 Md (Q8/FP8)
Modalités—
Fenêtre de contexte131 072
KV cache / token (Q8)17.9 ko/token
Paramètres totaux5.1 B
Paramètres actifs (session)5.12 B
Source des paramètres actifsactifs = total (modèle dense, config HF (google/gemma-4-E2B-it)) ; total : config Hugging Face (google/gemma-4-E2B-it)
Curation des paramètres actifs2026-11-04
Prix par session—
Tokens / tâche (Data Players)1.2 M
Tokens / tâche (Terminal-Bench)83.3 M (est.)
Ratio Terminal-Bench / Data Players69.4×
Coût / tâche (Terminal-Bench)—
Tokens d'entraînement—
Coupure de connaissance2025-01-01
Créé le2026-05-07T16:53:30.239713Z
Mis à jour le2026-11-04T06:04:46.923341Z
SourceLLM Stats
URLhttps://llm-stats.com/models/gemma-4-e2b-it

Prix par fournisseur

Statut du modèle (offres + auto-hébergement) : partiel — auto-hébergement DGX Spark (≤ 107 Md).

Toutes sources confondues : catalogue llm-stats, routeurs EUrouter/Cortecs et catalogues directs (OVHcloud, Scaleway, Infomaniak). Les prix sont convertis en USD pour la comparaison ; le prix publié et le taux BCE utilisé sont dans l'infobulle. Les prix inconnus restent indiqués « — ».

Aucune donnée.

Souveraineté qualifiée par offre (modèle × fournisseur) ; survolez le badge pour les critères.

Souveraineté

Scores par catégorie (top_scores)

CatégorieScore
language0.50
legal0.60
long_context0.20
math0.50
multimodal0.40
physics0.40
reasoning0.40
finance0.60
general0.40
healthcare0.40
agents0.30
biology0.40
chemistry0.40
tool_calling0.30
vision0.40

Scores par benchmark

BenchmarkCatégorieScoreMaxNormalisé (API)VérifiéAuto-déclaréDaté du
aime-2026math0.381.000.38nonoui2026-11-04T06:04:46.933074Z
big-bench-extra-hardreasoning0.221.000.22nonoui2026-11-04T06:04:46.953168Z
gpqareasoning0.431.000.43nonoui2026-11-04T06:04:46.940578Z
livecodebench-v6reasoning0.441.000.44nonoui2026-11-04T06:04:46.973017Z
mathvisionmath0.521.000.52nonoui2026-11-04T06:04:46.993127Z
medxpertqahealthcare0.231.000.23nonoui2026-11-04T06:04:47.013171Z
mmlu-prolanguage0.601.000.60nonoui2026-11-04T06:04:46.928352Z
mmmlulanguage0.671.000.67nonoui2026-11-04T06:04:46.963135Z
mmmu-provision0.441.000.44nonoui2026-11-04T06:04:46.983014Z
mrcr-v2-(8-needle)long_context0.191.000.19nonoui2026-11-04T06:04:47.003161Z
t2-benchagents0.291.000.29nonoui2026-11-04T06:04:47.023869Z

Classements

CatégorieRangScoreNote conservatriceBenchmarks évaluésMéthodeDaté duProvenance
agents196—-3.661trueskill2026-10-19model_history

Lecture : Note conservatrice = borne basse TrueSkill (niveau estimé − marge d’incertitude) ; Benchmarks évalués = nombre de benchmarks de la catégorie pris en compte. À performances égales, un modèle évalué sur moins de benchmarks obtient une note plus basse : la note ne se déduit pas d’un seul score. Provenance : rankings = top 50 de /rankings ; model_history = historique du modèle (/models/{id}/history), utilisé pour les rangs au-delà du top 50.

Données

Dates de curation (récupération depuis l'API) : models 2026-11-04 · benchmarks 2026-11-04 · scores 2026-11-04 · updates 2026-11-04 · account 2026-11-04 · model_history 2026-11-04 · rankings 2026-11-04