K2 Horizon 0.9B

Fiche modèle · Institute of Foundation Models · k2-horizon-0.9b

K2 Horizon 0.9B is the compact dense member of IFM's (MBZUAI) fully open K2 Horizon family: a 0.9B-class decoder-only reasoning model (28 layers, smaller 64K vocabulary; about 1.08B parameters in the released safetensors) pretrained on 5T tokens and post-trained with multi-teacher distillation from math/code, STEM, and instruction-following experts followed by on-policy distillation. Supports English and Chinese and a 131,072-token (128K) context window via YaRN RoPE scaling. Text in, text out, with per-request reasoning effort (high recommended) and JSON/XML tool-call formats. Released 2026-09-03 under Apache 2.0 with BF16 weights plus a GGUF build, an optional Uno discrete-diffusion LoRA adapter for faster decoding, intermediate checkpoints, and training code. No priced provider is cataloged yet.

Identité

OrganisationInstitute of Foundation Models
Pays—
Famille—
Typellm
Date de sortie2026-09-03
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)28.7 ko/token
Paramètres totaux900 M
Paramètres actifs (session)1.08 B
Source des paramètres actifsactifs = total (modèle dense, config HF (IFM/K2-Horizon-0.9B)) ; total : config Hugging Face (IFM/K2-Horizon-0.9B)
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'entraînement5 000 000 000 000
Coupure de connaissance—
Créé le2026-10-29T18:18:04.807088Z
Mis à jour le2026-11-04T06:04:48.083262Z
SourceLLM Stats
URLhttps://llm-stats.com/models/k2-horizon-0.9b

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
math0.40
physics0.30
reasoning0.50
general0.40
agents0.30
biology0.30
chemistry0.30
tool_calling0.30

Scores par benchmark

BenchmarkCatégorieScoreMaxNormalisé (API)VérifiéAuto-déclaréDaté du
aime-2025math0.421.000.42nonoui2026-11-04T06:04:48.087154Z
aime-2026math0.481.000.48nonoui2026-11-04T06:04:48.093047Z
bfcl-v4tool_calling0.281.000.28nonoui2026-11-04T06:04:48.143051Z
gpqareasoning0.271.000.27nonoui2026-11-04T06:04:48.113135Z
hmmt-feb-26math0.261.000.26nonoui2026-11-04T06:04:48.103123Z
humaneval+reasoning0.801.000.80nonoui2026-11-04T06:04:48.123357Z
livecodebench-v6reasoning0.371.000.37nonoui2026-11-04T06:04:48.139730Z
mbpp+reasoning0.681.000.68nonoui2026-11-04T06:04:48.129646Z

Classements

CatégorieRangScoreNote conservatriceBenchmarks évaluésMéthodeDaté duProvenance
agents218—-5.311trueskill2026-11-02model_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