SWE-2

Fiche modèle · Cognition · swe-2

SWE-2 is Cognition's most advanced coding model, post-trained from Moonshot AI's 2.8-trillion-parameter Kimi K3 with reinforcement learning scaled to the multi-trillion-parameter regime for the first time. A single RL run trains all reasoning-effort levels (medium, high, max) together with per-effort cost penalties tuned to the base model's Pareto frontier, advancing the whole cost-performance curve rather than a single operating point. On FrontierCode 1.1 Main it scores 50.0%, within one point of Claude Fable 5.1 (50.9%) while Cognition states it is 64% cheaper; on DeepSWE 1.1 it scores 73.0%; on Terminal-Bench 2.1 it scores 92.8%; on Terminal-Bench 4.0 it scores 27.3%. SWE-2 is available starting 2026-09-10 in Devin Desktop and Devin CLI, with Devin Web and Fusion rollout in progress; Cognition has not published a standalone public API, model card, or context-window specification independent of the Devin product.

Identité

OrganisationCognition
Pays—
Famille—
Typellm
Date de sortie2026-09-10
Licenceproprietary
Licence commercialenon
Licence open sourcenon
Poids ouvertsnon
Auto-hébergement (DGX Spark)non — poids fermés
Modalités—
Fenêtre de contexte—
KV cache / token (Q8)—
Paramètres totaux—
Paramètres actifs (session)—
Source des paramètres actifs—
Curation des paramètres actifs—
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 connaissance—
Créé le2026-10-29T18:18:29.070565Z
Mis à jour le2026-11-04T06:05:14.996263Z
SourceLLM Stats
URLhttps://llm-stats.com/models/swe-2

Prix par fournisseur

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
reasoning0.60
general0.60
agents0.60
code0.60
tool_calling0.60

Scores par benchmark

BenchmarkCatégorieScoreMaxNormalisé (API)VérifiéAuto-déclaréDaté du
deepswe-1.1code0.731.000.73nonoui2026-11-04T06:05:15.002952Z
frontiercode-1.1reasoning0.501.000.50nonoui2026-11-04T06:05:14.999566Z
terminal-bench-2.1reasoning0.931.000.93nonoui2026-11-04T06:05:15.006439Z
terminal-bench-4.0reasoning0.271.000.27nonoui2026-11-04T06:05:15.009742Z

Classements

CatégorieRangScoreNote conservatriceBenchmarks évaluésMéthodeDaté duProvenance
agents270.6035.654trueskill2026-11-04T08:08:47.839150Zrankings
code260.6038.664trueskill2026-11-04T08:08:39.742311Zrankings
reasoning370.6045.503trueskill2026-11-04T08:08:41.093494Zrankings

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