K2 Horizon MoVA 36B-A4B
K2 Horizon MoVA 36B-A4B is the sparse mid-size member of IFM's (MBZUAI) fully open K2 Horizon family: a Mixture-of-Experts reasoning model with 36B total and about 4B active parameters per token (48 layers, 100 routed experts with 8 active plus 1 shared expert) that also uses IFM's new Mixture-of-Value Attention (MoVA), where a router selects a few value-projection experts per token. Pretrained on 22.9T tokens with a native 524,288-token (512K) context window from midtraining onward. 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 FP8 and GGUF builds, intermediate checkpoints, training data, code, and logs. No priced provider is cataloged yet.
Identité
| Organisation | Institute of Foundation Models |
|---|---|
| Pays | — |
| Famille | — |
| Type | llm |
| Date de sortie | 2026-09-03 |
| Licence | apache 2.0 |
| Licence commerciale | oui |
| Licence open source | oui |
| Poids ouverts | oui |
| Auto-hébergement (DGX Spark) | oui — ≤ 107 Md (Q8/FP8) |
| Modalités | — |
| Fenêtre de contexte | 524 288 |
| KV cache / token (Q8) | 98.3 ko/token |
| Paramètres totaux | 36 B |
| Paramètres actifs (session) | 4 B |
| Source des paramètres actifs | actifs : nom (heuristique) ; total : config Hugging Face (IFM/K2-Horizon-MoVA-36B-A4B) |
| Curation des paramètres actifs | 2026-11-04 |
| Prix par session | — |
| Tokens / tâche (Data Players) | 1.2 M |
| Tokens / tâche (Terminal-Bench) | 83.3 M (est.) |
| Ratio Terminal-Bench / Data Players | 69.4× |
| Coût / tâche (Terminal-Bench) | — |
| Tokens d'entraînement | 22 900 000 000 000 |
| Coupure de connaissance | — |
| Créé le | 2026-10-29T18:18:05.247373Z |
| Mis à jour le | 2026-11-04T06:04:48.529777Z |
| Source | LLM Stats |
| URL | https://llm-stats.com/models/k2-horizon-mova-36b-a4b |
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.
SouverainetéScores par catégorie (top_scores)
| Catégorie | Score |
|---|---|
| knowledge | 0.40 |
| long_context | 0.70 |
| math | 0.20 |
| physics | 0.60 |
| reasoning | 0.40 |
| science | 0.40 |
| general | 0.40 |
| agents | 0.40 |
| biology | 0.60 |
| chemistry | 0.60 |
| code | 0.50 |
| tool_calling | 0.40 |
| vision | 0.30 |
Scores par benchmark
| Benchmark | Catégorie | Score | Max | Normalisé (API) | Vérifié | Auto-déclaré | Daté du |
|---|---|---|---|---|---|---|---|
| aa-lcr | reasoning | 0.66 | 1.00 | 0.66 | non | oui | 2026-11-04T06:04:48.570282Z |
| critpt | reasoning | 0.02 | 1.00 | 0.02 | non | oui | 2026-11-04T06:04:48.563023Z |
| gpqa | reasoning | 0.81 | 1.00 | 0.81 | non | oui | 2026-11-04T06:04:48.558893Z |
| humanity's-last-exam | reasoning | 0.25 | 1.00 | 0.25 | non | oui | 2026-11-04T06:04:48.553102Z |
| omniscience | science | 0.19 | 1.00 | 0.19 | non | oui | 2026-11-04T06:04:48.583067Z |
| omniscience-non-hallucination-rate | knowledge | 0.69 | 1.00 | 0.69 | non | oui | 2026-11-04T06:04:48.593173Z |
| scicode | reasoning | 0.39 | 1.00 | 0.39 | non | oui | 2026-11-04T06:04:48.542952Z |
| tau3-banking | agents | 0.27 | 1.00 | 0.27 | non | oui | 2026-11-04T06:04:48.533259Z |
| terminal-bench-2.1 | reasoning | 0.59 | 1.00 | 0.59 | non | oui | 2026-11-04T06:04:48.539821Z |
Classements
| Catégorie | Rang | Score | Note conservatrice | Benchmarks évalués | Méthode | Daté du | Provenance |
|---|---|---|---|---|---|---|---|
| knowledge | 19 | 0.40 | 13.74 | 2 | trueskill | 2026-11-04T08:08:45.139626Z | rankings |
| agents | 116 | — | 14.53 | 2 | trueskill | 2026-11-02 | model_history |
| code | 163 | — | 13.86 | 2 | trueskill | 2026-11-02 | model_history |
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