{"article_url":"https://newsverum.com/en/news/cohere-study-finds-agent-team-planning-efficiency","block":{"columns":[{"key":"team","label":{"ar":"تكوين الفريق","de":"Teamkonfiguration","en":"Team configuration","es":"Configuración del equipo","fr":"Configuration de l’équipe","pt":"Configuração da equipe","tr":"Ekip yapılandırması"},"language":"en","type":"text"},{"key":"reward","label":{"ar":"درجة المكافأة","de":"Belohnungswert","en":"Reward score","es":"Puntuación de recompensa","fr":"Score de récompense","pt":"Pontuação de recompensa","tr":"Ödül puanı"},"type":"number","unit":"percent"},{"key":"cost","label":{"ar":"إجمالي تكلفة الاستدلال (دولار أمريكي)","de":"Gesamte Inferenzkosten (USD)","en":"Total inference cost (USD)","es":"Coste total de inferencia (USD)","fr":"Coût total d’inférence (USD)","pt":"Custo total de inferência (USD)","tr":"Toplam çıkarım maliyeti (USD)"},"type":"number","unit":"USD"},{"key":"trial_design","label":{"ar":"تصميم التجربة","de":"Versuchsaufbau","en":"Trial design","es":"Diseño del ensayo","fr":"Plan d’essai","pt":"Desenho do ensaio","tr":"Deneme tasarımı"},"language":"en","type":"text"}],"description":{"ar":"نتائج ProgramBench من تقييم Cohere لفرق وكلاء البرمجة؛ عُرضت المكافأة وتكلفة الاستدلال كما وردتا في المصدر.","de":"ProgramBench-Ergebnisse aus Cohere-Bewertung von Programmieragenten-Teams; Belohnung und Inferenzkosten sind laut Quelle angegeben.","en":"ProgramBench results from Cohere’s coding-agent team evaluation; reward and inference cost are reported as stated by the source.","es":"Resultados de ProgramBench de la evaluación de equipos de agentes de programación de Cohere; recompensa y coste se presentan según la fuente.","fr":"Résultats ProgramBench de l’évaluation des équipes d’agents de programmation par Cohere ; score et coût sont rapportés selon la source.","pt":"Resultados do ProgramBench na avaliação de equipes de agentes de programação da Cohere; recompensa e custo são informados conforme a fonte.","tr":"Cohere’nin kodlama ajanı ekiplerini değerlendirdiği ProgramBench sonuçları; ödül puanı ve çıkarım maliyeti kaynakta raporlandığı biçimdedir."},"id":"programbench-team-results","methodology":{"ar":"المكافأة هي متوسط حصة الاختبارات السلوكية المخفية التي اجتازتها الفرق عبر 200 مهمة من ProgramBench. يمثل الإعدادان الأولان غير المتزامنين من GPT-5.6 Sol متوسط ثلاث جولات؛ وشُغّل الإعداد الثابت الذي يضم 16 عاملاً من GPT-5.6 Luna مرة واحدة. النتائج من بيئة اختبار Cohere.","de":"Belohnung ist der mittlere Anteil bestandener verdeckter Verhaltenstests über 200 ProgramBench-Aufgaben. Die ersten beiden asynchronen GPT-5.6-Sol-Konfigurationen mitteln drei Durchläufe; die feste Konfiguration mit 16 GPT-5.6-Luna-Agenten wurde einmal ausgeführt. Ergebnisse aus Cohere-eigenem Testsystem.","en":"Reward is the mean share of hidden behavioral tests passed across 200 ProgramBench tasks. The first two asynchronous GPT-5.6 Sol configurations average three runs; the fixed configuration with 16 GPT-5.6 Luna workers was run once. Results are from Cohere’s own test harness.","es":"La recompensa es la proporción media de pruebas de comportamiento ocultas superadas en 200 tareas de ProgramBench. Las dos primeras configuraciones asíncronas GPT-5.6 Sol promedian tres ejecuciones; la configuración fija con 16 agentes GPT-5.6 Luna se ejecutó una vez. Resultados del entorno de pruebas de Cohere.","fr":"La récompense est la part moyenne des tests comportementaux cachés réussis sur 200 tâches ProgramBench. Les deux premières configurations asynchrones GPT-5.6 Sol sont des moyennes de trois exécutions ; la configuration fixe avec 16 agents GPT-5.6 Luna n’a été exécutée qu’une fois. Résultats issus du banc d’essai de Cohere.","pt":"A recompensa é a proporção média de testes comportamentais ocultos aprovados em 200 tarefas do ProgramBench. As duas primeiras configurações assíncronas GPT-5.6 Sol são médias de três execuções; a configuração fixa com 16 agentes GPT-5.6 Luna foi executada uma vez. Resultados do ambiente de testes da Cohere.","tr":"Ödül, 200 ProgramBench görevi boyunca gizli davranış testlerinden geçen payın ortalamasıdır. İlk iki eşzamansız GPT-5.6 Sol yapılandırması üç çalıştırmanın ortalamasıdır; sabit 16 GPT-5.6 Luna çalışanı yapılandırması yalnızca bir kez çalıştırılmıştır. Sonuçlar Cohere’nin kendi test düzeneğine aittir."},"rowTranslations":[{"team":{"ar":"GPT-5.6 Sol؛ قائد غير متزامن يفوض التخطيط فقط","de":"GPT-5.6 Sol; asynchroner Leiter delegiert nur die Planung","en":"GPT-5.6 Sol; async lead delegates planning only","es":"GPT-5.6 Sol; líder asíncrono delega solo la planificación","fr":"GPT-5.6 Sol ; responsable asynchrone ne délègue que la planification","pt":"GPT-5.6 Sol; líder assíncrono delega apenas o planejamento","tr":"GPT-5.6 Sol; eşzamansız lider yalnızca planlamayı devrediyor"},"trial_design":{"ar":"200 مهمة؛ متوسط 3 جولات","de":"200 Aufgaben; Mittelwert aus 3 Durchläufen","en":"200 tasks; mean of 3 runs","es":"200 tareas; promedio de 3 ejecuciones","fr":"200 tâches ; moyenne de 3 exécutions","pt":"200 tarefas; média de 3 execuções","tr":"200 görev; 3 çalıştırmanın ortalaması"}},{"team":{"ar":"GPT-5.6 Sol؛ قائد غير متزامن يفوض جميع المراحل","de":"GPT-5.6 Sol; asynchroner Leiter delegiert alle Phasen","en":"GPT-5.6 Sol; async lead delegates all stages","es":"GPT-5.6 Sol; líder asíncrono delega todas las etapas","fr":"GPT-5.6 Sol ; responsable asynchrone délègue toutes les étapes","pt":"GPT-5.6 Sol; líder assíncrono delega todas as etapas","tr":"GPT-5.6 Sol; eşzamansız lider tüm aşamaları devrediyor"},"trial_design":{"ar":"200 مهمة؛ متوسط 3 جولات","de":"200 Aufgaben; Mittelwert aus 3 Durchläufen","en":"200 tasks; mean of 3 runs","es":"200 tareas; promedio de 3 ejecuciones","fr":"200 tâches ; moyenne de 3 exécutions","pt":"200 tarefas; média de 3 execuções","tr":"200 görev; 3 çalıştırmanın ortalaması"}},{"team":{"ar":"قائد GPT-5.6 Sol؛ و16 وكيلاً للتخطيط من GPT-5.6 Luna","de":"GPT-5.6-Sol-Leiter; 16 GPT-5.6-Luna-Planungsagenten","en":"GPT-5.6 Sol lead; 16 GPT-5.6 Luna planning workers","es":"Líder GPT-5.6 Sol; 16 agentes planificadores GPT-5.6 Luna","fr":"Responsable GPT-5.6 Sol ; 16 agents de planification GPT-5.6 Luna","pt":"Líder GPT-5.6 Sol; 16 agentes planejadores GPT-5.6 Luna","tr":"GPT-5.6 Sol lider; 16 GPT-5.6 Luna planlama ajanı"},"trial_design":{"ar":"تكوين ثابت من 16 وكيلاً؛ جولة واحدة","de":"Feste Konfiguration mit 16 Agenten; 1 Durchlauf","en":"Fixed 16-worker configuration; 1 run","es":"Configuración fija de 16 agentes; 1 ejecución","fr":"Configuration fixe à 16 agents ; 1 exécution","pt":"Configuração fixa de 16 agentes; 1 execução","tr":"Sabit 16 ajanlı yapılandırma; 1 çalıştırma"}}],"rows":[{"cost":249,"reward":63.6,"team":"GPT-5.6 Sol; async lead delegates planning only","trial_design":"200 tasks; mean of 3 runs"},{"cost":510,"reward":63.2,"team":"GPT-5.6 Sol; async lead delegates all stages","trial_design":"200 tasks; mean of 3 runs"},{"cost":155,"reward":61.8,"team":"GPT-5.6 Sol lead; 16 GPT-5.6 Luna planning workers","trial_design":"Fixed 16-worker configuration; 1 run"}],"sourceUrls":["https://cohere.com/blog/lessons-from-scaling-teams-of-frontier-coding-agents"],"title":{"ar":"نتائج فرق Cohere في ProgramBench","de":"Cohere-Teamresultate auf ProgramBench","en":"Cohere ProgramBench team results","es":"Resultados de equipos de Cohere en ProgramBench","fr":"Résultats des équipes Cohere sur ProgramBench","pt":"Resultados das equipes da Cohere no ProgramBench","tr":"Cohere ProgramBench ekip sonuçları"},"type":"table","updatedAt":"2026-10-15T16:48:43Z"},"version":1}