Solar Mini 4

Fiche modèle · Upstage · solar-mini4

Solar Mini 4 is Upstage's cost-efficient proprietary MoE language model (35B total / 3B active) for agentic use where speed and cost matter. API id solar-mini4 (version solar-mini4-260922). Text→text with English, Korean, and Japanese; reasoning mode, tool calling, and structured outputs. Upstage Console docs: 512K context, up to 128K max output, February 2026 training cutoff. List pricing $0.10 input / $0.40 output / $0.01 cached per 1M tokens. Independent Artificial Analysis benches (not vendor self-report): AA-LCR 83.3, SciCode 47.6, HLE text-only 25.8, Terminal-Bench 4.0 1.0, GDPval-AA v2.1 1072 Elo, AA-Briefcase v1.1 872 Elo. No separate Upstage-published bench table found beyond console/product claims; AA rows stored as third-party.

Identité

OrganisationUpstage
Pays—
Famille—
Typellm
Date de sortie2026-09-22
Licenceproprietary
Licence commercialenon
Licence open sourcenon
Poids ouvertsnon
Auto-hébergement (DGX Spark)non — poids fermés
Modalitéstext
Fenêtre de contexte512 000
KV cache / token (Q8)—
Paramètres totaux35 B
Paramètres actifs (session)3 B
Source des paramètres actifsweb : https://console.upstage.ai/docs/models/solar-mini-4
Curation des paramètres actifs2026-10-19
Prix par session0.18 $ (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)1.51 $ (est.)
Tokens d'entraînement—
Coupure de connaissance2026-02-01
Créé le2026-10-07T17:32:43.796296Z
Mis à jour le2026-11-04T06:05:12.952899Z
SourceLLM Stats
URLhttps://llm-stats.com/models/solar-mini4

Prix par fournisseur

Statut du modèle (offres + auto-hébergement) : non.

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
Upstagellm-statsnon0.10 $0.40 $————active

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

Souveraineté

Scores par catégorie (top_scores)

CatégorieScore
long_context0.80
math0.40
physics0.50
productivity872.00
reasoning324.30
general324.30
agents648.00
biology0.50
chemistry0.50
code0.20
tool_calling0.00

Scores par benchmark

BenchmarkCatégorieScoreMaxNormalisé (API)VérifiéAuto-déclaréDaté du
aa-briefcase-1.1agents872.003 000.000.29nonnon2026-11-04T06:05:12.971139Z
aa-lcrreasoning0.831.000.83nonnon2026-11-04T06:05:12.956030Z
gdpval-aa-2.1general1 072.003 000.000.36nonnon2026-11-04T06:05:12.968181Z
humanity's-last-exam-(no-tools,-text-only)reasoning0.261.000.26nonnon2026-11-04T06:05:12.961931Z
scicodereasoning0.481.000.48nonnon2026-11-04T06:05:12.959002Z
terminal-bench-4.0reasoning0.011.000.01nonnon2026-11-04T06:05:12.964980Z

Classements

CatégorieRangScoreNote conservatriceBenchmarks évaluésMéthodeDaté duProvenance
agents180—2.643trueskill2026-10-19model_history
code135—17.172trueskill2026-10-19model_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