{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-23T07:21:26.498Z","headline":"Poolside 发布 Laguna S 2.1：118B 参数 MoE 模型，长程编码能力对标更大模型","description":"Poolside 发布 Laguna S 2.1，一个 118B 总参数、8B 激活参数的 MoE 模型，支持 1M token 上下文窗口，从训练到发布用时不到九周。在 Terminal-Bench 2.1 上以 70.2% 的得分超越多数同尺寸模型，在 DeepSWE 上得分 40.4%。模型权重已开源，完整评估轨迹可在 trajectories.poolside.ai 获取。","url":"https://www.aioga.com/news/cmrv4r5hv00hhbizaqm90z102/","mainEntityOfPage":"https://www.aioga.com/news/cmrv4r5hv00hhbizaqm90z102/","datePublished":"2026-07-21T20:43:27.067Z","dateModified":"2026-07-21T20:43:27.067Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://poolside.ai/blog/introducing-laguna-s-2-1","https://aihot.virxact.com/items/cmrv4r5hv00hhbizaqm90z102"],"canonicalUrl":"https://www.aioga.com/news/cmrv4r5hv00hhbizaqm90z102/","directAnswer":{"@type":"Answer","text":"Poolside 发布 Laguna S 2.1。该模型采用 MoE 架构，总参数为 118B，每个 token 激活 8B 参数，支持最高 1M token 上下文窗口，模型权重已经开源。","url":"https://www.aioga.com/news/cmrv4r5hv00hhbizaqm90z102/","dateCreated":"2026-07-21T20:43:27.067Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"poolside.ai source article","url":"https://poolside.ai/blog/introducing-laguna-s-2-1","datePublished":"2026-07-21T20:43:27.067Z","provider":{"@type":"Organization","name":"poolside.ai","url":"https://poolside.ai/blog/introducing-laguna-s-2-1"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmrv4r5hv00hhbizaqm90z102","datePublished":"2026-07-21T20:43:27.067Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmrv4r5hv00hhbizaqm90z102"}}],"aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","originalPublisher":{"name":"poolside.ai","url":"https://poolside.ai/blog/introducing-laguna-s-2-1"},"article":{"id":"cmrv4r5hv00hhbizaqm90z102","slug":"cmrv4r5hv00hhbizaqm90z102","url":"https://www.aioga.com/news/cmrv4r5hv00hhbizaqm90z102/","title":"Poolside 发布 Laguna S 2.1：118B 参数 MoE 模型，长程编码能力对标更大模型","title_en":"Laguna S 2.1","summary":"Poolside 发布 Laguna S 2.1，一个 118B 总参数、8B 激活参数的 MoE 模型，支持 1M token 上下文窗口，从训练到发布用时不到九周。在 Terminal-Bench 2.1 上以 70.2% 的得分超越多数同尺寸模型，在 DeepSWE 上得分 40.4%。模型权重已开源，完整评估轨迹可在 trajectories.poolside.ai 获取。","source":"Hacker News 热门（buzzing.cc 中文翻译）","sourceUrl":"https://poolside.ai/blog/introducing-laguna-s-2-1","aiHotUrl":"https://aihot.virxact.com/items/cmrv4r5hv00hhbizaqm90z102","publishedAt":"2026-07-21T20:43:27.067Z","category":"模型更新","score":78,"selected":false,"articleBody":["Today we’re releasing Laguna S 2.1, a significant step forward in our development of models that pursue longer horizon work and make effective use of reasoning.","Laguna S 2.1 is a 118B total parameter Mixture-of-Experts (MoE) model with 8B activated parameters per token and supports a context window of up to 1M tokens in thinking and no-thinking modes. It went from the start of training to launch in under nine weeks, and on long-horizon coding benchmarks it holds its own against models many times its size. For every benchmark score we publish today, we are releasing full trajectories for every trial in the final evaluation set at trajectories.poolside.ai ：http://trajectories.poolside.ai/ .","Laguna S 2.1 is, as far as we can measure, the most capable agentic coding model in its weight class by a wide margin .","S 2.1 scores 70.2% on Terminal-Bench 2.1 in our agent harness with thinking enabled. Its compact size makes it uniquely suitable for complex work on local machines.","A ranked comparison of reported Terminal-Bench 2.1 scores.","Terminal-Bench 2.1 evaluates a wide, high-quality set of long-horizon tasks where an agent model is connected to its environment through a terminal. Laguna S 2.1 is a standout model in its size category on this benchmark.","Scatter plot of disclosed total parameter count on a logarithmic axis against Terminal-Bench 2.1 score. Laguna S 2.1 is highlighted as a triangle."],"articleImages":[],"mediaStatus":"none","articleBodyZh":["今天我们发布了 Laguna S 2.1，这是我们在开发能够执行长远任务并有效利用推理的模型方面迈出的重要一步。","Laguna S 2.1 是一个拥有 118B 总参数的专家混合（Mixture-of-Experts, MoE）模型，每个 token 激活 8B 参数，并且在思考模式和非思考模式下支持最长 1M token 的上下文窗口。从训练开始到上线不到九周，在长远编码基准测试中，其表现可以与体积大数倍的模型相媲美。对于今天发布的每个基准分数，我们都在 trajectories.poolside.ai 发布了最终评估集每次试验的完整轨迹：http://trajectories.poolside.ai/。","截至目前可测量的结果，Laguna S 2.1 在其重量级别中是最有能力的自主编码模型，相差悬殊。","S 2.1 在我们的代理测试环境中启用思考功能时，在 Terminal-Bench 2.1 上得分为 70.2%。其紧凑的体积使其特别适合在本地机器上执行复杂工作。","Terminal-Bench 2.1 报告分数的排名比较。","Terminal-Bench 2.1 评估了一组广泛、高质量的长远任务，其中代理模型通过终端与其环境连接。Laguna S 2.1 在该基准测试的同类模型中表现出众。","在对数坐标轴上展示的已披露总参数数量与 Terminal-Bench 2.1 分数的散点图。Laguna S 2.1 以三角形突出显示。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Poolside 发布 Laguna S 2.1。该模型采用 MoE 架构，总参数为 118B，每个 token 激活 8B 参数，支持最高 1M token 上下文窗口，模型权重已经开源。","background":"正文称，该模型从开始训练到发布用时不到九周，面向更长周期的工作与推理场景。其 Terminal-Bench 2.1 得分为 70.2%，DeepSWE 得分为 40.4%。","viewpoint":"Aioga 判断，Laguna S 2.1 的主要看点是以较低激活参数量支持长上下文和代理式编码任务。不过，“同级最强”等表述来自发布方，仍需独立评测验证。","implications":"该发布可能提升市场对 MoE 模型在本地复杂编码任务中效率表现的关注。完整评估轨迹的公开也为外部复核提供了材料，但现有结果不能直接推及所有开发场景。","nextStep":"值得关注独立机构能否复现其 Terminal-Bench 2.1 与 DeepSWE 成绩，并比较不同硬件上的推理成本、运行稳定性及 1M token 上下文在实际项目中的有效利用情况。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-07-23T04:07:40.172Z","sourceHash":"a36bdb4b3ea249f8","review":{"approved":true,"groundedness":96,"clarity":92,"duplicationRisk":18,"blockingIssues":[],"notes":["“同级最强”被明确归因于发布方并提示需独立验证，处理恰当。","“可能提升市场关注”属于审慎表达的影响判断，并未冒充既成事实。","DeepSWE 40.4% 和模型权重已开源的信息来自来源摘要，而非所列正文摘录，但仍属于所提供来源材料。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","low-source-overlap","no-html","independent-ai-review"]}},"tags":["模型更新","Hacker News 热门（buzzing.cc 中文翻译）"],"translations":{"zh-CN":{"title":"Poolside 发布 Laguna S 2.1：118B 参数 MoE 模型，长程编码能力对标更大模型","summary":"Poolside 发布 Laguna S 2.1，一个 118B 总参数、8B 激活参数的 MoE 模型，支持 1M token 上下文窗口，从训练到发布用时不到九周。在 Terminal-Bench 2.1 上以 70.2% 的得分超越多数同尺寸模型，在 DeepSWE 上得分 40.4%。模型权重已开源，完整评估轨迹可在 trajectories.poolside.ai 获取。","category":"模型更新","source":"poolside.ai","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside 发布 Laguna S 2.1：118B 参数 MoE 模型，长程编码能力对标更大模型 - Aioga AI资讯","description":"Poolside 发布 Laguna S 2.1，一个 118B 总参数、8B 激活参数的 MoE 模型，支持 1M token 上下文窗口，从训练到发布用时不到九周。在 Terminal-Bench 2.1 上以 70.2% 的得分超越多数同尺寸模型，在 DeepSWE 上得分 40.4%。模型权重已开源，完整评估轨迹可在 trajectories.poo...","url":"https://www.aioga.com/news/cmrv4r5hv00hhbizaqm90z102/"},"en":{"title":"Poolside releases Laguna S 2.1: 118B parameter MoE model, long-range encoding capability comparable to larger models","summary":"Poolside released Laguna S 2.1, a MoE model with 118B total parameters and 8B active parameters, supporting a 1M token context window, taking less than nine weeks from training to release. It scored 70.2% on Terminal-Bench 2.1, surpassing most models of the same size, and scored 40.4% on DeepSWE. The model weights have been open-sourced, and the complete evaluation trajectory can be accessed at trajectories.poolside.ai.","category":"Models","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside releases Laguna S 2.1: 118B parameter MoE model, long-range encoding capability comparable to larger models - Aioga AI News","description":"Poolside released Laguna S 2.1, a MoE model with 118B total parameters and 8B active parameters, supporting a 1M token context window, taking less than nine weeks from training to...","url":"https://www.aioga.com/en/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:45:39.916Z"},"ja":{"title":"Poolside、Laguna S 2.1：118BパラメータMoEモデルを公開、長距離エンコーディング能力はより大きなモデルに匹敵","summary":"Poolside は Laguna S 2.1 をリリースしました。これは総パラメータ118B、アクティベーションパラメータ8BのMoEモデルで、1Mトークンのコンテキストウィンドウをサポートします。トレーニングからリリースまでにかかった時間は9週間未満です。Terminal-Bench 2.1ではスコア70.2%で同サイズの多くのモデルを超え、DeepSWEではスコア40.4%を達成しました。モデルの重みはオープンソース化されており、完全な評価トラジェクトリーはtrajectories.poolside.aiで入手可能です。","category":"モデル更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside、Laguna S 2.1：118BパラメータMoEモデルを公開、長距離エンコーディング能力はより大きなモデルに匹敵 - Aioga AIニュース","description":"Poolside は Laguna S 2.1 をリリースしました。これは総パラメータ118B、アクティベーションパラメータ8BのMoEモデルで、1Mトークンのコンテキストウィンドウをサポートします。トレーニングからリリースまでにかかった時間は9週間未満です。Terminal-Bench 2.1ではスコア70.2%で同サイズの多くのモデルを超え、DeepSW...","url":"https://www.aioga.com/ja/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:45:45.829Z"},"ko":{"title":"Poolside가 Laguna S 2.1 출시: 118B 파라미터 MoE 모델, 장거리 인코딩 능력이 더 큰 모델과 견줄 만함","summary":"Poolside는 Laguna S 2.1을 발표했습니다. 이 모델은 총 118B 파라미터와 8B 활성화 파라미터를 가진 MoE 모델로, 1M 토큰의 컨텍스트 윈도우를 지원하며, 훈련부터 발표까지 9주도 채 걸리지 않았습니다. Terminal-Bench 2.1에서 70.2% 점수로 대부분의 동급 모델을 능가했고, DeepSWE에서는 40.4% 점수를 기록했습니다. 모델 가중치는 오픈소스로 제공되며, 전체 평가 경로는 trajectories.poolside.ai에서 확인할 수 있습니다.","category":"모델 업데이트","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside가 Laguna S 2.1 출시: 118B 파라미터 MoE 모델, 장거리 인코딩 능력이 더 큰 모델과 견줄 만함 - Aioga AI 뉴스","description":"Poolside는 Laguna S 2.1을 발표했습니다. 이 모델은 총 118B 파라미터와 8B 활성화 파라미터를 가진 MoE 모델로, 1M 토큰의 컨텍스트 윈도우를 지원하며, 훈련부터 발표까지 9주도 채 걸리지 않았습니다. Terminal-Bench 2.1에서 70.2% 점수로 대부분의 동급 모델을 능가했고, DeepS...","url":"https://www.aioga.com/ko/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:46:41.362Z"},"es":{"title":"Poolside lanza Laguna S 2.1: modelo MoE con 118B parámetros, capacidad de codificación a largo plazo comparable a modelos más grandes","summary":"Poolside lanzó Laguna S 2.1, un modelo MoE con 118B parámetros totales y 8B parámetros de activación, que soporta una ventana de contexto de 1M tokens, y desde el entrenamiento hasta el lanzamiento tomó menos de nueve semanas. En Terminal-Bench 2.1 obtuvo una puntuación de 70,2%, superando a la mayoría de los modelos de tamaño similar, y en DeepSWE obtuvo un 40,4%. Los pesos del modelo se han hecho de código abierto, y la trayectoria completa de evaluación está disponible en trajectories.poolside.ai.","category":"Modelos","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside lanza Laguna S 2.1: modelo MoE con 118B parámetros, capacidad de codificación a largo plazo comparable a modelos más grandes - Aioga Noticias de IA","description":"Poolside lanzó Laguna S 2.1, un modelo MoE con 118B parámetros totales y 8B parámetros de activación, que soporta una ventana de contexto de 1M tokens, y desde el entrenamiento has...","url":"https://www.aioga.com/es/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:46:34.258Z"},"fr":{"title":"Poolside publie Laguna S 2.1 : modèle MoE à 118 milliards de paramètres, capacité de codage longue portée comparable à des modèles plus grands","summary":"Poolside a publié Laguna S 2.1, un modèle MoE avec 118 milliards de paramètres totaux et 8 milliards de paramètres d'activation, prenant en charge une fenêtre de contexte de 1 million de tokens, et allant de l'entraînement à la publication en moins de neuf semaines. Il dépasse la plupart des modèles de même taille avec un score de 70,2% sur Terminal-Bench 2.1, et obtient un score de 40,4% sur DeepSWE. Les poids du modèle ont été rendus publics, et la trajectoire complète de l'évaluation est disponible sur trajectories.poolside.ai.","category":"Modèles","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside publie Laguna S 2.1 : modèle MoE à 118 milliards de paramètres, capacité de codage longue portée comparable à des modèles plus grands - Aioga Actualités IA","description":"Poolside a publié Laguna S 2.1, un modèle MoE avec 118 milliards de paramètres totaux et 8 milliards de paramètres d'activation, prenant en charge une fenêtre de contexte de 1 mill...","url":"https://www.aioga.com/fr/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:47:30.167Z"},"de":{"title":"Poolside veröffentlicht Laguna S 2.1: 118B Parameter MoE-Modell, Langstrecken-Codierungsfähigkeit entspricht größeren Modellen","summary":"Poolside hat Laguna S 2.1 veröffentlicht, ein MoE-Modell mit insgesamt 118B Parametern und 8B Aktivierungsparametern, das ein Kontextfenster von 1M Token unterstützt und von Training bis Veröffentlichung weniger als neun Wochen benötigte. Auf Terminal-Bench 2.1 übertrifft es die meisten Modelle gleicher Größe mit einer Punktzahl von 70,2 % und erzielt auf DeepSWE eine Punktzahl von 40,4 %. Die Modellgewichte sind Open Source, und die vollständigen Bewertungsergebnisse können unter trajectories.poolside.ai eingesehen werden.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside veröffentlicht Laguna S 2.1: 118B Parameter MoE-Modell, Langstrecken-Codierungsfähigkeit entspricht größeren Modellen - Aioga KI-News","description":"Poolside hat Laguna S 2.1 veröffentlicht, ein MoE-Modell mit insgesamt 118B Parametern und 8B Aktivierungsparametern, das ein Kontextfenster von 1M Token unterstützt und von Traini...","url":"https://www.aioga.com/de/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:47:25.310Z"},"pt-BR":{"title":"Poolside lançou Laguna S 2.1: modelo MoE com 118B parâmetros, capacidade de codificação de longo alcance comparável a modelos maiores","summary":"Poolside lançou o Laguna S 2.1, um modelo MoE com 118 bilhões de parâmetros totais e 8 bilhões de parâmetros de ativação, suportando uma janela de contexto de 1 milhão de tokens, com menos de nove semanas desde o treinamento até o lançamento. No Terminal-Bench 2.1, atingiu 70,2% de pontuação, superando a maioria dos modelos do mesmo tamanho, e obteve 40,4% no DeepSWE. Os pesos do modelo foram abertos, e a trajetória completa de avaliação pode ser acessada em trajectories.poolside.ai.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside lançou Laguna S 2.1: modelo MoE com 118B parâmetros, capacidade de codificação de longo alcance comparável a modelos maiores - Aioga Notícias de IA","description":"Poolside lançou o Laguna S 2.1, um modelo MoE com 118 bilhões de parâmetros totais e 8 bilhões de parâmetros de ativação, suportando uma janela de contexto de 1 milhão de tokens, c...","url":"https://www.aioga.com/pt-BR/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:49:25.052Z"},"ru":{"title":"Poolside выпустил Laguna S 2.1: MoE модель с 118B параметрами, способность к длинным последовательностям сопоставима с более крупными моделями","summary":"Poolside выпустила Laguna S 2.1, модель MoE с общим числом параметров 118B и 8B параметров активации, поддерживающую контекстное окно в 1M токенов; от обучения до релиза прошло менее девяти недель. На Terminal-Bench 2.1 модель набрала 70,2%, превзойдя большинство моделей того же размера, а на DeepSWE результат составил 40,4%. Весы модели открыты, полный отчет о проверке доступен на trajectories.poolside.ai.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside выпустил Laguna S 2.1: MoE модель с 118B параметрами, способность к длинным последовательностям сопоставима с более крупными моделями - Aioga Новости ИИ","description":"Poolside выпустила Laguna S 2.1, модель MoE с общим числом параметров 118B и 8B параметров активации, поддерживающую контекстное окно в 1M токенов; от обучения до релиза прошло мен...","url":"https://www.aioga.com/ru/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:49:26.912Z"},"ar":{"title":"Poolside أصدرت Laguna S 2.1: نموذج MoE بمعاملات 118B، قدرة الترميز طويل المدى تقارن بالنماذج الأكبر","summary":"أطلقت Poolside نموذج Laguna S 2.1، وهو نموذج MoE يحتوي على 118 مليار معلمة إجمالية و8 مليارات معلمة تنشيط، ويدعم نافذة سياق تصل إلى 1 مليون رمز. استغرق من التدريب إلى الإطلاق أقل من تسعة أسابيع. تجاوز النموذج معظم النماذج ذات الحجم المماثل على Terminal-Bench 2.1 بدرجة 70.2٪، وحقق درجة 40.4٪ على DeepSWE. تم نشر أوزان النموذج كمصدر مفتوح، ويمكن الحصول على المسار الكامل للتقييم على trajectories.poolside.ai.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside أصدرت Laguna S 2.1: نموذج MoE بمعاملات 118B، قدرة الترميز طويل المدى تقارن بالنماذج الأكبر - Aioga أخبار الذكاء الاصطناعي","description":"أطلقت Poolside نموذج Laguna S 2.1، وهو نموذج MoE يحتوي على 118 مليار معلمة إجمالية و8 مليارات معلمة تنشيط، ويدعم نافذة سياق تصل إلى 1 مليون رمز. استغرق من التدريب إلى الإطلاق أقل م...","url":"https://www.aioga.com/ar/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:50:12.754Z"},"hi":{"title":"Poolside ने Laguna S 2.1 जारी किया: 118B पैरामीटर MoE मॉडल, लंबी क्रमिक कोडिंग क्षमता बड़ी मॉडलों के बराबर","summary":"Poolside ने Laguna S 2.1 जारी किया, एक 118B कुल पैरामीटर, 8B सक्रिय पैरामीटर वाला MoE मॉडल, जो 1M टोकन संदर्भ विंडो का समर्थन करता है, प्रशिक्षण से लेकर रिलीज़ तक नौ सप्ताह से कम समय में। Terminal-Bench 2.1 पर 70.2% के स्कोर के साथ अधिकांश समान आकार के मॉडलों को पीछे छोड़ दिया, और DeepSWE पर 40.4% का स्कोर हासिल किया। मॉडल वेट्स खुले स्रोत में उपलब्ध हैं, पूर्ण मूल्यांकन ट्रैसेज trajectories.poolside.ai पर प्राप्त किए जा सकते हैं।","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside ने Laguna S 2.1 जारी किया: 118B पैरामीटर MoE मॉडल, लंबी क्रमिक कोडिंग क्षमता बड़ी मॉडलों के बराबर - Aioga AI समाचार","description":"Poolside ने Laguna S 2.1 जारी किया, एक 118B कुल पैरामीटर, 8B सक्रिय पैरामीटर वाला MoE मॉडल, जो 1M टोकन संदर्भ विंडो का समर्थन करता है, प्रशिक्षण से लेकर रिलीज़ तक नौ सप्ताह से कम स...","url":"https://www.aioga.com/hi/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:50:15.562Z"},"it":{"title":"Poolside rilascia Laguna S 2.1: modello MoE con 118B parametri, la capacità di codifica a lungo raggio è equivalente a modelli più grandi","summary":"Poolside ha rilasciato Laguna S 2.1, un modello MoE con 118 miliardi di parametri totali e 8 miliardi di parametri attivi, che supporta una finestra di contesto di 1 milione di token, impiegando meno di nove settimane dalla fase di addestramento al rilascio. Su Terminal-Bench 2.1 ha superato la maggior parte dei modelli di dimensioni simili con un punteggio del 70,2%, e su DeepSWE ha ottenuto un punteggio del 40,4%. I pesi del modello sono stati resi open source e la traiettoria di valutazione completa è disponibile su trajectories.poolside.ai.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside rilascia Laguna S 2.1: modello MoE con 118B parametri, la capacità di codifica a lungo raggio è equivalente a modelli più grandi - Aioga Notizie IA","description":"Poolside ha rilasciato Laguna S 2.1, un modello MoE con 118 miliardi di parametri totali e 8 miliardi di parametri attivi, che supporta una finestra di contesto di 1 milione di tok...","url":"https://www.aioga.com/it/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:51:05.273Z"},"nl":{"title":"Poolside brengt Laguna S 2.1 uit: 118B parameters MoE-model, lange-afstand coderingcapaciteit vergelijkbaar met grotere modellen","summary":"Poolside heeft Laguna S 2.1 uitgebracht, een MoE-model met 118 miljard totale parameters en 8 miljard activatieparameters, dat een contextvenster van 1 miljoen tokens ondersteunt. Van training tot release duurde het minder dan negen weken. Het model behaalde een score van 70,2% op Terminal-Bench 2.1 en overtrof daarmee de meeste modellen van hetzelfde formaat, en op DeepSWE werd een score van 40,4% behaald. De modelgewichten zijn open source beschikbaar en het volledige evaluatietraject is te vinden op trajectories.poolside.ai.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside brengt Laguna S 2.1 uit: 118B parameters MoE-model, lange-afstand coderingcapaciteit vergelijkbaar met grotere modellen - Aioga AI-nieuws","description":"Poolside heeft Laguna S 2.1 uitgebracht, een MoE-model met 118 miljard totale parameters en 8 miljard activatieparameters, dat een contextvenster van 1 miljoen tokens ondersteunt....","url":"https://www.aioga.com/nl/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:51:04.503Z"},"tr":{"title":"Poolside, Laguna S 2.1: 118B parametreli MoE modeli yayınladı, uzun menzilli kodlama yeteneği daha büyük modellerle karşılaştırılıyor","summary":"Poolside, 118B toplam parametre ve 8B aktivasyon parametresine sahip bir MoE modeli olan Laguna S 2.1'i duyurdu, 1M token bağlam penceresini destekliyor ve eğitimden yayına kadar dokuz haftadan kısa sürdü. Terminal-Bench 2.1'de %70,2 skorla aynı boyuttaki çoğu modeli geride bırakırken, DeepSWE'de %40,4 puan aldı. Model ağırlıkları açık kaynaklıdır ve tam değerlendirme izleri trajectories.poolside.ai adresinde bulunabilir.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside, Laguna S 2.1: 118B parametreli MoE modeli yayınladı, uzun menzilli kodlama yeteneği daha büyük modellerle karşılaştırılıyor - Aioga AI Haberleri","description":"Poolside, 118B toplam parametre ve 8B aktivasyon parametresine sahip bir MoE modeli olan Laguna S 2.1'i duyurdu, 1M token bağlam penceresini destekliyor ve eğitimden yayına kadar d...","url":"https://www.aioga.com/tr/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:51:48.714Z"},"vi":{"title":"Poolside phát hành Laguna S 2.1: mô hình MoE 118B tham số, khả năng mã hóa dài hạn so sánh với các mô hình lớn hơn","summary":"Poolside phát hành Laguna S 2.1, một mô hình MoE với tổng 118 tỷ tham số và 8 tỷ tham số kích hoạt, hỗ trợ cửa sổ ngữ cảnh 1 triệu token, từ huấn luyện đến phát hành chỉ mất chưa đến chín tuần. Trên Terminal-Bench 2.1, đạt 70,2% điểm, vượt qua hầu hết các mô hình cùng kích thước, trên DeepSWE đạt 40,4%. Trọng số mô hình đã được mở nguồn, toàn bộ lộ trình đánh giá có thể truy cập tại trajectories.poolside.ai.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside phát hành Laguna S 2.1: mô hình MoE 118B tham số, khả năng mã hóa dài hạn so sánh với các mô hình lớn hơn - Tin tức AI Aioga","description":"Poolside phát hành Laguna S 2.1, một mô hình MoE với tổng 118 tỷ tham số và 8 tỷ tham số kích hoạt, hỗ trợ cửa sổ ngữ cảnh 1 triệu token, từ huấn luyện đến phát hành chỉ mất chưa đ...","url":"https://www.aioga.com/vi/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:51:51.002Z"},"id":{"title":"Poolside merilis Laguna S 2.1: model MoE 118B parameter, kemampuan pengkodean jangka panjang setara dengan model yang lebih besar","summary":"Poolside merilis Laguna S 2.1, sebuah model MoE dengan total 118B parameter dan 8B parameter aktivasi, mendukung jendela konteks 1M token, dari pelatihan hingga rilis memakan waktu kurang dari sembilan minggu. Di Terminal-Bench 2.1, model ini mencapai skor 70,2%, melampaui sebagian besar model dengan ukuran serupa, dan memperoleh skor 40,4% di DeepSWE. Bobot model telah dibuka untuk umum, dan jalur evaluasi lengkap dapat diakses di trajectories.poolside.ai.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside merilis Laguna S 2.1: model MoE 118B parameter, kemampuan pengkodean jangka panjang setara dengan model yang lebih besar - Berita AI Aioga","description":"Poolside merilis Laguna S 2.1, sebuah model MoE dengan total 118B parameter dan 8B parameter aktivasi, mendukung jendela konteks 1M token, dari pelatihan hingga rilis memakan waktu...","url":"https://www.aioga.com/id/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:52:31.840Z"},"th":{"title":"Poolside เปิดตัว Laguna S 2.1: โมเดล MoE 118B พารามิเตอร์ ความสามารถในการเข้ารหัสระยะยาวเทียบเท่ากับโมเดลที่ใหญ่กว่า","summary":"Poolside เปิดตัว Laguna S 2.1 ซึ่งเป็นโมเดล MoE ที่มีพารามิเตอร์ทั้งหมด 118B และพารามิเตอร์แอกทีฟ 8B รองรับหน้าต่างบริบท 1M token ใช้เวลาน้อยกว่าเก้าสัปดาห์ตั้งแต่การฝึกจนถึงการเปิดตัว ใน Terminal-Bench 2.1 ทำคะแนนได้ 70.2% แซงหน้าส่วนใหญ่ของโมเดลที่มีขนาดเดียวกัน ใน DeepSWE ทำคะแนนได้ 40.4% น้ำหนักของโมเดลถูกเปิดเผยสู่สาธารณะ และเส้นทางการประเมินผลทั้งหมดสามารถดูได้ที่ trajectories.poolside.ai","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside เปิดตัว Laguna S 2.1: โมเดล MoE 118B พารามิเตอร์ ความสามารถในการเข้ารหัสระยะยาวเทียบเท่ากับโมเดลที่ใหญ่กว่า - ข่าว AI Aioga","description":"Poolside เปิดตัว Laguna S 2.1 ซึ่งเป็นโมเดล MoE ที่มีพารามิเตอร์ทั้งหมด 118B และพารามิเตอร์แอกทีฟ 8B รองรับหน้าต่างบริบท 1M token ใช้เวลาน้อยกว่าเก้าสัปดาห์ตั้งแต่การฝึกจนถึงการเปิ...","url":"https://www.aioga.com/th/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:52:40.095Z"},"pl":{"title":"Poolside wydaje Laguna S 2.1: model MoE z 118B parametrów, zdolność kodowania długiego zasięgu porównywalna z większymi modelami","summary":"Poolside wydało Laguna S 2.1, model MoE z łączną liczbą parametrów 118B i 8B parametrów aktywacji, wspierający okno kontekstu o długości 1M tokenów, którego czas od treningu do publikacji wyniósł mniej niż dziewięć tygodni. W Terminal-Bench 2.1 uzyskał wynik 70,2%, przewyższając większość modeli o podobnej wielkości, a na DeepSWE osiągnął wynik 40,4%. Wagi modelu zostały udostępnione publicznie, a pełna ścieżka oceny jest dostępna na trajectories.poolside.ai.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Poolside wydaje Laguna S 2.1: model MoE z 118B parametrów, zdolność kodowania długiego zasięgu porównywalna z większymi modelami - Aioga Wiadomości AI","description":"Poolside wydało Laguna S 2.1, model MoE z łączną liczbą parametrów 118B i 8B parametrów aktywacji, wspierający okno kontekstu o długości 1M tokenów, którego czas od treningu do pub...","url":"https://www.aioga.com/pl/news/cmrv4r5hv00hhbizaqm90z102/","contentTranslated":true,"sourceHash":"970445e9c22f7a7d","translatedAt":"2026-07-22T17:53:29.175Z"}}}}