IBM Granite 4.2 3B

Fiche modèle · IBM · granite-4.2-3b

IBM Granite 4.2 3B is the compact dense decoder-only reasoning LLM in the Granite 4.2 family. It supports native think, non-think, and low-effort modes, tool calling, and a 128K native context window with extension to 512K. Unlike the 8B and 30B, it did not receive the extra agentic RL block. Open weights on Hugging Face under Apache 2.0. Text-only (not multimodal).

Identité

OrganisationIBM
Pays—
Famille—
Typellm
Date de sortie2026-08-25
Licenceapache 2.0
Licence commercialeoui
Licence open sourceoui
Poids ouvertsoui
Auto-hébergement (DGX Spark)oui — ≤ 107 Md (Q8/FP8)
Modalitéstext
Fenêtre de contexte131 072
KV cache / token (Q8)41.0 ko/token
Paramètres totaux3 B
Paramètres actifs (session)3.66 B
Source des paramètres actifsactifs = total (modèle dense, config HF (ibm-granite/granite-4.2-3b)) ; total : config Hugging Face (ibm-granite/granite-4.2-3b)
Curation des paramètres actifs2026-11-04
Prix par session0.05 $ (1 000 000 tokens entrée + 200 000 sortie, prix le moins cher par fournisseur)
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)0.45 $ (est.)
Tokens d'entraînement—
Coupure de connaissance—
Créé le2026-08-30T23:49:06.642023Z
Mis à jour le2026-11-04T06:04:47.833274Z
SourceLLM Stats
URLhttps://llm-stats.com/models/granite-4.2-3b

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 « — ».

FournisseurSourceSouverainetéEntrée (/M tok)Sortie (/M tok)QuantificationContexteLatence (s)Débit (tok/s)Statut
DeepInfrallm-statsnon0.03 $0.12 $————active

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

Souveraineté

Scores par catégorie (top_scores)

CatégorieScore
instruction_following0.70
knowledge0.30
language0.70
legal0.70
long_context0.60
math0.60
physics0.40
reasoning0.50
finance0.70
general0.60
healthcare0.70
agents0.40
biology0.40
chemistry0.40
code0.20
creativity0.30
tool_calling0.50
writing0.30

Scores par benchmark

BenchmarkCatégorieScoreMaxNormalisé (API)VérifiéAuto-déclaréDaté du
aime-2025math0.781.000.78nonoui2026-11-04T06:04:47.863128Z
arena-hard-v2general0.351.000.35nonoui2026-11-04T06:04:47.913005Z
bfcl-v4tool_calling0.521.000.52nonoui2026-11-04T06:04:47.843045Z
gpqareasoning0.551.000.55nonoui2026-11-04T06:04:47.883037Z
hmmt25math0.671.000.67nonoui2026-11-04T06:04:47.873103Z
ifbenchinstruction_following0.741.000.74nonoui2026-11-04T06:04:47.921173Z
livecodebench-v6reasoning0.701.000.70nonoui2026-11-04T06:04:47.887614Z
mmlu-prolanguage0.681.000.68nonoui2026-11-04T06:04:47.903071Z
profbenchreasoning0.321.000.32nonoui2026-11-04T06:04:47.853101Z
ruler-128klong_context0.551.000.55nonoui2026-11-04T06:04:47.937582Z
ruler-64klong_context0.681.000.68nonoui2026-11-04T06:04:47.933209Z
scicodereasoning0.241.000.24nonoui2026-11-04T06:04:47.892927Z
tau3-benchagents0.461.000.46nonoui2026-11-04T06:04:47.837685Z

Classements

CatégorieRangScoreNote conservatriceBenchmarks évaluésMéthodeDaté duProvenance
knowledge340.30-1.611trueskill2026-11-04T08:08:45.139626Zrankings
agents186—1.103trueskill2026-10-19model_history
code277—-7.181trueskill2026-10-19model_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