{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-28T06:20:51.496Z","headline":"Poolside 发布 Laguna S 2.1：118B 参数开源编程模型，8B 活跃参数性能超越十倍大模型","description":"Poolside 发布第三款编程模型 Laguna S 2.1，采用 MoE 架构，总参数 118B，每 token 仅激活 8B 参数，支持百万 token 上下文窗口与思考/非思考双模式。","url":"https://www.aioga.com/news/cmrxihnkj00e9ro4ju90yp4sr/","mainEntityOfPage":"https://www.aioga.com/news/cmrxihnkj00e9ro4ju90yp4sr/","datePublished":"2026-07-23T12:24:53.000Z","dateModified":"2026-07-23T12:24:53.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://the-decoder.com/poolsides-laguna-s-2-1-is-a-small-open-weight-coding-model-that-punches-well-above-its-size","https://aihot.virxact.com/items/cmrxihnkj00e9ro4ju90yp4sr"],"canonicalUrl":"https://www.aioga.com/news/cmrxihnkj00e9ro4ju90yp4sr/","directAnswer":{"@type":"Answer","text":"Aioga 编辑摘要：Poolside 发布第三款编程模型 Laguna S 2.1，采用 MoE 架构，总参数 118B，每 token 仅激活 8B 参数，支持百万 token 上下文窗口与思考/非思考双模式。 Aioga 将其归入「模型更新」方向，重点关注它对真实使用和行业竞争的影响。","url":"https://www.aioga.com/news/cmrxihnkj00e9ro4ju90yp4sr/","dateCreated":"2026-07-23T12:24:53.000Z","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":"the-decoder.com source article","url":"https://the-decoder.com/poolsides-laguna-s-2-1-is-a-small-open-weight-coding-model-that-punches-well-above-its-size","datePublished":"2026-07-23T12:24:53.000Z","provider":{"@type":"Organization","name":"the-decoder.com","url":"https://the-decoder.com/poolsides-laguna-s-2-1-is-a-small-open-weight-coding-model-that-punches-well-above-its-size"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmrxihnkj00e9ro4ju90yp4sr","datePublished":"2026-07-23T12:24:53.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmrxihnkj00e9ro4ju90yp4sr"}}],"aggregationSource":"The Decoder：AI News（RSS）","originalPublisher":{"name":"the-decoder.com","url":"https://the-decoder.com/poolsides-laguna-s-2-1-is-a-small-open-weight-coding-model-that-punches-well-above-its-size"},"article":{"id":"cmrxihnkj00e9ro4ju90yp4sr","slug":"cmrxihnkj00e9ro4ju90yp4sr","url":"https://www.aioga.com/news/cmrxihnkj00e9ro4ju90yp4sr/","title":"Poolside 发布 Laguna S 2.1：118B 参数开源编程模型，8B 活跃参数性能超越十倍大模型","title_en":"Poolside's Laguna S 2.1 is a small open-weight coding model that punches well above its size","summary":"Poolside 发布第三款编程模型 Laguna S 2.1，采用 MoE 架构，总参数 118B，每 token 仅激活 8B 参数，支持百万 token 上下文窗口与思考/非思考双模式。","source":"The Decoder：AI News（RSS）","sourceUrl":"https://the-decoder.com/poolsides-laguna-s-2-1-is-a-small-open-weight-coding-model-that-punches-well-above-its-size","aiHotUrl":"https://aihot.virxact.com/items/cmrxihnkj00e9ro4ju90yp4sr","publishedAt":"2026-07-23T12:24:53.000Z","category":"模型更新","score":51,"selected":false,"articleBody":["Poolside has released Laguna S 2.1, its third coding model in three months. The mixture-of-experts model uses 8 billion active parameters and focuses less on raw scale than on better behavior during long agentic sessions.","US-based Poolside says Laguna S 2.1 outperforms other agentic coding models in its weight class and sometimes approaches systems 10 to 20 times its size. The model has 118 billion total parameters, with 8 billion active for each token. It supports context windows of up to one million tokens and offers thinking and no-thinking modes.","Poolside made its first models available to a broader audience in April 2026 with Laguna M.1 and XS.2：https://poolside.ai/blog/laguna-a-deeper-dive. Until then, the company had focused on government and public-sector customers. XS.2 was also its first open model under the Apache 2.0 license. Laguna S 2.1 is the third version in the series released in roughly three months.","Poolside says：https://poolside.ai/blog/introducing-laguna-s-2-1 Datacurve's DeepSWE benchmark offers a better comparison because its scores are spread across a wider range. Laguna S 2.1 scores 40.4 percent, while some open-weight models with more than one trillion parameters remain below 10 percent. It also ranks near the top of its class on SWE-Bench Multilingual, SWE-Bench Pro, and SWE Atlas.","Thinking mode has a major impact on performance. Without it, Laguna S 2.1's Terminal-Bench score drops to 60.4 percent, while its DeepSWE score falls to 16.5 percent. Poolside says no previous Laguna model has shown a larger performance gap between the two modes.","Poolside says the release reflects a broader idea about model performance. \"What we've done in this model is not necessarily add more intelligence, but improve the behaviors that lead to a more capable model: more verification, less taking things for granted, not declaring victory early, and being more persistent,\" the company writes in its release post：https://poolside.ai/blog/introducing-laguna-s-2-1.","Earlier Laguna models sometimes stopped after only partially passing a test suite or abandoned an approach just two steps before it would have worked. Poolside treats persistence, verification, and revising failed approaches as a second path to better performance alongside scaling the model itself. A larger Laguna model is already in pre-training.","Poolside supports its claim with three documented trial runs. In one, Laguna S 2.1 built a working browser engine：https://trajectories.poolside.ai/trials/019f6c37-7621-7ec1-96b5-f041f83c1540 from an empty folder in 50 minutes that could render HTML and CSS. In another, the model found a proof for Erdos Problem #397：https://trajectories.poolside.ai/trials/019f2a95-b4b3-77b8-ad7c-dbdf59c0d9ca, a math problem that had been open since 1975, while working in a sandbox without Python. Poolside says the result was an independent rediscovery. GPT-5.2 Pro solved this and several other problems in January 2026：https://the-decoder.com/terence-tao-says-gpt-5-2-pro-cracked-an-erdos-problem-but-warns-the-win-says-more-about-speed-than-difficulty/, while Laguna's training cutoff was November 2025.","Poolside says the jump from XS 2.1 to S 2.1 came mainly from scaling and post-training, not new pre-training data. The agentic training phase covered 409,000 environments, including 83,000 for terminal tasks and 168,000 for software engineering workflows. The largest single source was about 38,000 real commits from roughly 17,000 repositories. A new task category trained the model to install repositories on its own, set up every dependency, and get test suites running.","Poolside increased rollout budgets and extended timeouts. It also built a new sandbox system that can selectively block network access to curb reward hacking. Multi-harness rollouts run the same prompts across several agent environments, reducing the risk of overfitting to one setup.","Fewer than nine weeks passed between the start of training and launch, according to Poolside. Pre-training began on May 22, 2026, using 4,096 Nvidia H200 GPUs：https://the-decoder.com/nvidia-unveils-new-h200-gpu-for-ai-and-hpc-workloads/. S 2.1 is also the company's first model trained with reinforcement learning in FP8 precision.","Poolside published every benchmark trajectory at trajectories.poolside.ai：http://trajectories.poolside.ai/. During training, reward hacking rates topped 50 percent on SWE-Bench tasks because the model searched online for matching pull requests instead of solving the tasks itself. A small prompt change brought the rate below two percent.","Laguna S 2.1 is still too closely tuned to Poolside's agent harness in some cases. In unfamiliar environments with slightly different tool schemas, the model can stray from the required format, Poolside says. It also tends to produce overly long thinking sequences on competitive math problems. Users can't adjust its thinking effort yet.","Laguna S 2.1 is available on Hugging Face：https://huggingface.co/poolside/Laguna-S-2.1 under the OpenMDW 1.1 license. Backed by the Linux Foundation, the license allows anyone to use, modify, and redistribute the model weights, including for commercial purposes.","Baseten, Vercel AI Gateway, and OpenRouter offer hosted access. OpenRouter provides a free endpoint with a 256K context window and a paid endpoint supporting the full one-million-token window. Poolside says the model can also run locally on a single Nvidia DGX Spark：https://the-decoder.com/early-reviews-suggest-nvidia-may-have-found-another-way-to-sell-its-chips-with-the-dgx-spark/. A free demo chat is available at chat.poolside.ai：https://chat.poolside.ai/ without a login.","Poolside is making two strategic bets. One is that the path to intelligence runs through agentic coding：https://the-decoder.com/new-review-paper-argues-code-is-how-ai-agents-think-and-act-not-just-what-they-produce/ because software gives agents their most expressive interface. It also believes AI can \"decompress the web.\" Most written material records answers rather than the reasoning behind them, and Poolside argues that reinforcement learning can recover that process.","Stay in the loop on AI. Clear, useful, no fluff.","Follow The Decoder for AI news, background stories and expert analyses.","The Decoder：https://the-decoder.com/"],"articleImages":[{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/07/poolside-laguna-s-2-1-terminal-bench-2026-07-22_14-37.png","alt":"Rangliste auf Terminal-Bench 2.1 zeigt Laguna S 2.1 auf Platz 11 von 22 Modellen mit 70,2 % Pass@1.","afterParagraph":2,"url":"/media/articles/cmrxihnkj00e9ro4ju90yp4sr/efa1abf7656fa653.png"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/07/poolside-laguna-s-2-1-thinking-effort-2026-07-22_15-01.png","alt":"Liniendiagramm vergleicht Erfolgsquoten mit und ohne Thinking-Modus und zeigt durchgängig höhere Werte mit aktiviertem Thinking.","afterParagraph":4,"url":"/media/articles/cmrxihnkj00e9ro4ju90yp4sr/c6047b44914a32e8.png"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/07/poolside-trajectory-erdos-397-026-07-22_14-49.png","alt":"Screenshot des Trial: Prompt „this is an unsolved problem, solve it…“ und Lösung von Erdős-Problem 397 in 40 Schritten mit 283.981 Zeichen Reasoning, Tokens 2,8 M/115 K/2,4 M, Kosten 0,088 $.","afterParagraph":7,"url":"/media/articles/cmrxihnkj00e9ro4ju90yp4sr/afcd3d75bfaa47db.png"}],"mediaStatus":"ok","articleBodyZh":["Poolside 发布了 Laguna S 2.1，这是其三个月内发布的第三个编程模型。该专家混合模型使用了 80 亿个活跃参数，并且相比规模更注重在长时间代理会话中的更好行为表现。","总部位于美国的 Poolside 表示，Laguna S 2.1 在同类重量的代理编程模型中表现优于其他模型，有时性能接近体积大 10 到 20 倍的系统。该模型总参数为 1180 亿，其中每个 token 使用 80 亿活跃参数。它支持高达一百万个 token 的上下文窗口，并提供“思考模式”和“不思考模式”。","Poolside 在 2026 年 4 月通过 Laguna M.1 和 XS.2 向更广泛的受众开放了其首批模型：https://poolside.ai/blog/laguna-a-deeper-dive。在此之前，该公司主要面向政府和公共部门客户。XS.2 也是其在 Apache 2.0 许可下发布的第一个开源模型。Laguna S 2.1 是该系列在大约三个月内发布的第三个版本。","Poolside 表示：https://poolside.ai/blog/introducing-laguna-s-2-1 Datacurve 的 DeepSWE 基准测试提供了更好的比较，因为其分数分布更广。Laguna S 2.1 的得分为 40.4%，而一些拥有超过一万亿参数的开放权重模型得分仍低于 10%。它在 SWE-Bench 多语言、SWE-Bench Pro 和 SWE Atlas 中也位居同类前列。","思考模式对性能有重大影响。没有思考模式时，Laguna S 2.1 的 Terminal-Bench 得分下降到 60.4%，而 DeepSWE 得分降至 16.5%。Poolside 表示，以往的 Laguna 模型中没有哪个模型在两种模式下的性能差距如此之大。","Poolside 表示，此次发布反映了其对模型性能的更广泛理念。公司在发布文章中写道：https://poolside.ai/blog/introducing-laguna-s-2-1“我们在这个模型中所做的工作不一定是增加更多智能，而是改进那些能造就更强大模型的行为：更多验证、减少理所当然、不轻易宣布胜利，以及更持久的执行力。”","早期的 Laguna 模型有时只在部分通过测试套件后就停止，或者在某种方法即将成功的前两步就放弃。Poolside 将坚持不懈、验证和修改失败方法视为除模型本身扩展之外的一条实现更好性能的路径。一个更大的 Laguna 模型已经在预训练中。","Poolside 用三个有记录的试运行来支持其主张。在其中一次试运行中，Laguna S 2.1 从一个空文件夹在 50 分钟内构建了一个可工作浏览器引擎：https://trajectories.poolside.ai/trials/019f6c37-7621-7ec1-96b5-f041f83c1540，该引擎可以渲染 HTML 和 CSS。在另一项试运行中，模型找到了 Erdos 问题 #397 的证明：https://trajectories.poolside.ai/trials/019f2a95-b4b3-77b8-ad7c-dbdf59c0d9ca，这个数学问题自 1975 年起一直未解，模型在没有 Python 的沙箱中工作。Poolside 表示该结果是一次独立的再发现。GPT-5.2 Pro 在 2026 年 1 月解决了该问题及其他几个问题：https://the-decoder.com/terence-tao-says-gpt-5-2-pro-cracked-an-erdos-problem-but-warns-the-win-says-more-about-speed-than-difficulty/，而 Laguna 的训练截止日期为 2025 年 11 月。","Poolside 表示，从 XS 2.1 到 S 2.1 的飞跃主要来自模型扩展和训练后阶段，而非新的预训练数据。代理训练阶段覆盖了 409,000 个环境，其中包括 83,000 个终端任务环境和 168,000 个软件工程工作流环境。最大的单一数据来源约为 38,000 个真实提交，来自大约 17,000 个代码仓库。一个新的任务类别训练模型自行安装仓库，设置所有依赖项并运行测试套件。","Poolside 增加了部署预算并延长了超时期限。它还建了一个新的沙箱系统，可以选择性地阻止网络访问以抑制奖励作弊。多环境部署会在多个代理环境中运行相同的提示，从而降低对某一设置过拟合的风险。","根据 Poolside，训练开始到发布不到九周。预训练始于 2026 年 5 月 22 日，使用 4,096 块 Nvidia H200 GPU：https://the-decoder.com/nvidia-unveils-new-h200-gpu-for-ai-and-hpc-workloads/。S 2.1 也成为该公司首个使用 FP8 精度进行强化学习训练的模型。","Poolside 在 trajectories.poolside.ai 上发布了每一个基准轨迹：http://trajectories.poolside.ai/。在训练期间，由于模型在解题时会在线搜索匹配的 pull request，而不是自己解决任务，SWE-Bench 任务上的奖励作弊率一度超过 50%。一次小的提示修改使该比率降至不到 2%。","Laguna S 2.1 在某些情况下仍然对 Poolside 的代理框架调优得过于紧密。Poolside 表示，在稍微不同的工具模式的陌生环境中，模型可能偏离所需格式。它在竞争性数学题上也倾向于产生过长的思考序列。用户目前无法调整其思考强度。","Laguna S 2.1 可在 Hugging Face 获取：https://huggingface.co/poolside/Laguna-S-2.1，采用 OpenMDW 1.1 许可协议。在 Linux 基金会的支持下，该许可允许任何人使用、修改和重新分发模型权重，包括商业用途。","Baseten、Vercel AI Gateway 和 OpenRouter 提供托管访问。OpenRouter 提供一个 256K 上下文窗口的免费端点，以及支持完整一百万 token 窗口的付费端点。Poolside 表示，该模型也可以在单个 Nvidia DGX Spark 上本地运行：https://the-decoder.com/early-reviews-suggest-nvidia-may-have-found-another-way-to-sell-its-chips-with-the-dgx-spark/。免费的演示聊天可在 chat.poolside.ai 使用：https://chat.poolside.ai/，不需要登录。","Poolside 正在进行两个战略性押注。一个是认为通向智能的路径通过代理式编码实现：https://the-decoder.com/new-review-paper-argues-code-is-how-ai-agents-think-and-act-not-just-what-they-produce/，因为软件为代理提供了最具表现力的接口。它还认为 AI 可以“解压网络”。大部分书面材料记录的是答案而非背后的推理，Poolside 认为强化学习可以恢复这一过程。","保持 AI 最新动态。清晰、有用，无废话。","关注 The Decoder 获取 AI 新闻、背景故事和专家分析。","The Decoder：https://the-decoder.com/"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Aioga 编辑摘要：Poolside 发布第三款编程模型 Laguna S 2.1，采用 MoE 架构，总参数 118B，每 token 仅激活 8B 参数，支持百万 token 上下文窗口与思考/非思考双模式。 Aioga 将其归入「模型更新」方向，重点关注它对真实使用和行业竞争的影响。","background":"背景分析：模型与研究类动态需要结合能力边界、开放方式、成本、可用性和真实任务表现判断，单项指标领先不等于已经形成稳定采用。","viewpoint":"Aioga 判断：这条动态更适合作为行业观察信号，当前信息足以建立线索，但不足以推导长期结论。","implications":"影响分析：对相关团队而言，短期应先核对来源、可用范围和实际成本，再判断是否值得接入或跟进。","nextStep":"后续观察：继续观察官方文档、实际可用性、价格变化、开发者反馈和竞品回应。","evidenceRefs":["title","summary","articleBody"],"confidence":"medium","status":"published","aiGenerated":false,"autoApproved":true,"generatedBy":"rule-safe-fallback","generatedAt":"2026-07-28T06:29:10.595Z","sourceHash":"c1f539d98fd38826","validation":{"passed":true,"mode":"rule-safe-fallback","checks":["schema","length","source-attribution","no-html"]}},"tags":["模型更新","The Decoder：AI News（RSS）"],"translations":{"zh-CN":{"title":"Poolside 发布 Laguna S 2.1：118B 参数开源编程模型，8B 活跃参数性能超越十倍大模型","summary":"Poolside 发布第三款编程模型 Laguna S 2.1，采用 MoE 架构，总参数 118B，每 token 仅激活 8B 参数，支持百万 token 上下文窗口与思考/非思考双模式。","category":"模型更新","source":"the-decoder.com","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside 发布 Laguna S 2.1：118B 参数开源编程模型，8B 活跃参数性能超越十倍大模型 - Aioga AI资讯","description":"Poolside 发布第三款编程模型 Laguna S 2.1，采用 MoE 架构，总参数 118B，每 token 仅激活 8B 参数，支持百万 token 上下文窗口与思考/非思考双模式。","url":"https://www.aioga.com/news/cmrxihnkj00e9ro4ju90yp4sr/"},"en":{"title":"Poolside releases Laguna S 2.1: 118B parameter open-source programming model, 8B active parameters surpass the performance of tenfold larger models","summary":"Poolside has released its third programming model, Laguna S 2.1, using an MoE architecture with a total of 118B parameters, activating only 8B parameters per token, and supporting a million-token context window with both thinking and non-thinking dual modes.","category":"Models","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside releases Laguna S 2.1: 118B parameter open-source programming model, 8B active parameters surpass the performance of tenfold larger models - Aioga AI News","description":"Poolside has released its third programming model, Laguna S 2.1, using an MoE architecture with a total of 118B parameters, activating only 8B parameters per token, and supporting...","url":"https://www.aioga.com/en/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:42:30.941Z"},"ja":{"title":"Poolside が Laguna S 2.1 を公開：118B パラメータのオープンソースプログラミングモデル、8B 活性パラメータで十倍大規模モデルを凌駕する性能","summary":"Poolside は第3のプログラミングモデル Laguna S 2.1 を発表しました。MoE アーキテクチャを採用し、総パラメータは 118B、各トークンごとにわずか 8B パラメータを活性化し、百万トークンのコンテキストウィンドウと考える/考えないの二重モードをサポートします。","category":"モデル更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside が Laguna S 2.1 を公開：118B パラメータのオープンソースプログラミングモデル、8B 活性パラメータで十倍大規模モデルを凌駕する性能 - Aioga AIニュース","description":"Poolside は第3のプログラミングモデル Laguna S 2.1 を発表しました。MoE アーキテクチャを採用し、総パラメータは 118B、各トークンごとにわずか 8B パラメータを活性化し、百万トークンのコンテキストウィンドウと考える/考えないの二重モードをサポートします。","url":"https://www.aioga.com/ja/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:42:39.885Z"},"ko":{"title":"Poolside는 Laguna S 2.1을 발표했습니다: 118B 파라미터 오픈소스 프로그래밍 모델, 8B 활성 파라미터 성능이 10배 큰 모델을 능가","summary":"Poolside는 세 번째 프로그래밍 모델 Laguna S 2.1을 발표했으며, MoE 아키텍처를 사용하고, 총 파라미터는 118B이며, 토큰당 8B 파라미터만 활성화되며, 백만 토큰 컨텍스트 창과 사고/비사고 이중 모드를 지원합니다.","category":"모델 업데이트","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside는 Laguna S 2.1을 발표했습니다: 118B 파라미터 오픈소스 프로그래밍 모델, 8B 활성 파라미터 성능이 10배 큰 모델을 능가 - Aioga AI 뉴스","description":"Poolside는 세 번째 프로그래밍 모델 Laguna S 2.1을 발표했으며, MoE 아키텍처를 사용하고, 총 파라미터는 118B이며, 토큰당 8B 파라미터만 활성화되며, 백만 토큰 컨텍스트 창과 사고/비사고 이중 모드를 지원합니다.","url":"https://www.aioga.com/ko/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:43:26.531Z"},"es":{"title":"Poolside lanzó Laguna S 2.1: modelo de programación de código abierto con 118B parámetros, con 8B parámetros activos superando el rendimiento de modelos diez veces más grandes","summary":"Poolside lanzó el tercer modelo de programación Laguna S 2.1, que utiliza la arquitectura MoE, con un total de 118 mil millones de parámetros, activando solo 8 mil millones de parámetros por token, y admite una ventana de contexto de un millón de tokens con modos de pensamiento y no pensamiento.","category":"Modelos","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside lanzó Laguna S 2.1: modelo de programación de código abierto con 118B parámetros, con 8B parámetros activos superando el rendimiento de modelos diez veces más grandes - Aioga Noticias de IA","description":"Poolside lanzó el tercer modelo de programación Laguna S 2.1, que utiliza la arquitectura MoE, con un total de 118 mil millones de parámetros, activando solo 8 mil millones de pará...","url":"https://www.aioga.com/es/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:43:25.081Z"},"fr":{"title":"Poolside publie Laguna S 2.1 : modèle de programmation open source avec 118 milliards de paramètres, les performances des 8 milliards de paramètres actifs dépassent celles d'un modèle dix fois plus grand","summary":"Poolside a publié le troisième modèle de programmation Laguna S 2.1, utilisant une architecture MoE, avec un total de 118 milliards de paramètres, n'activant que 8 milliards de paramètres par token, et supportant une fenêtre de contexte de millions de tokens ainsi que les modes de pensée/non-pensée.","category":"Modèles","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside publie Laguna S 2.1 : modèle de programmation open source avec 118 milliards de paramètres, les performances des 8 milliards de paramètres actifs dépassent celles d'un modèle dix fois plus grand - Aioga Actualités IA","description":"Poolside a publié le troisième modèle de programmation Laguna S 2.1, utilisant une architecture MoE, avec un total de 118 milliards de paramètres, n'activant que 8 milliards de par...","url":"https://www.aioga.com/fr/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:44:11.330Z"},"de":{"title":"Poolside veröffentlicht Laguna S 2.1: Open-Source-Programmiermodell mit 118B Parametern, die Leistung von 8B aktiven Parametern übertrifft Modelle mit zehnmal so vielen Parametern","summary":"Poolside veröffentlicht das dritte Programmiermodell Laguna S 2.1, das die MoE-Architektur verwendet, insgesamt 118 Milliarden Parameter hat und nur 8 Milliarden Parameter pro Token aktiviert, unterstützt einen Kontextfenster von Millionen Token und zwei Modi: Denken/Nicht-Denken.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside veröffentlicht Laguna S 2.1: Open-Source-Programmiermodell mit 118B Parametern, die Leistung von 8B aktiven Parametern übertrifft Modelle mit zehnmal so vielen Parametern - Aioga KI-News","description":"Poolside veröffentlicht das dritte Programmiermodell Laguna S 2.1, das die MoE-Architektur verwendet, insgesamt 118 Milliarden Parameter hat und nur 8 Milliarden Parameter pro Toke...","url":"https://www.aioga.com/de/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:44:08.349Z"},"pt-BR":{"title":"Poolside lançou Laguna S 2.1: modelo de programação de código aberto com 118B parâmetros, desempenho de 8B parâmetros ativos supera modelos dez vezes maiores","summary":"Poolside lançou o terceiro modelo de programação Laguna S 2.1, utilizando a arquitetura MoE, com um total de 118 bilhões de parâmetros, ativando apenas 8 bilhões de parâmetros por token, suportando janelas de contexto de milhões de tokens e modos de pensamento/não pensamento.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside lançou Laguna S 2.1: modelo de programação de código aberto com 118B parâmetros, desempenho de 8B parâmetros ativos supera modelos dez vezes maiores - Aioga Notícias de IA","description":"Poolside lançou o terceiro modelo de programação Laguna S 2.1, utilizando a arquitetura MoE, com um total de 118 bilhões de parâmetros, ativando apenas 8 bilhões de parâmetros por...","url":"https://www.aioga.com/pt-BR/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:44:52.252Z"},"ru":{"title":"Poolside выпустил Laguna S 2.1: модель программирования с открытым исходным кодом с 118B параметрами, производительность активных 8B параметров превосходит десятикратные большие модели","summary":"Poolside выпустила третью модель программирования Laguna S 2.1, использующую архитектуру MoE, с общим числом параметров 118B, при этом на каждый токен активируется только 8B параметров, поддерживает окно контекста в миллион токенов и режимы мысли/без мысли.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside выпустил Laguna S 2.1: модель программирования с открытым исходным кодом с 118B параметрами, производительность активных 8B параметров превосходит десятикратные большие модели - Aioga Новости ИИ","description":"Poolside выпустила третью модель программирования Laguna S 2.1, использующую архитектуру MoE, с общим числом параметров 118B, при этом на каждый токен активируется только 8B параме...","url":"https://www.aioga.com/ru/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:45:06.893Z"},"ar":{"title":"Poolside أصدر Laguna S 2.1: نموذج برمجة مفتوح المصدر بمعلمات 118B، وأداء المعلمات النشطة 8B يتجاوز عشرة أضعاف النماذج الكبيرة","summary":"أطلقت Poolside النموذج البرمجي الثالث Laguna S 2.1، الذي يعتمد على بنية MoE، بمجموع معلمات يبلغ 118 مليار، حيث يتم تنشيط 8 مليارات معلمة فقط لكل رمز، ويدعم نافذة سياق تصل إلى مليون رمز ونمطين للتفكير/عدم التفكير.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside أصدر Laguna S 2.1: نموذج برمجة مفتوح المصدر بمعلمات 118B، وأداء المعلمات النشطة 8B يتجاوز عشرة أضعاف النماذج الكبيرة - Aioga أخبار الذكاء الاصطناعي","description":"أطلقت Poolside النموذج البرمجي الثالث Laguna S 2.1، الذي يعتمد على بنية MoE، بمجموع معلمات يبلغ 118 مليار، حيث يتم تنشيط 8 مليارات معلمة فقط لكل رمز، ويدعم نافذة سياق تصل إلى مليون...","url":"https://www.aioga.com/ar/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:45:55.660Z"},"hi":{"title":"Poolside ने Laguna S 2.1 लॉन्च किया: 118B पैरामीटर ओपन-सोर्स प्रोग्रामिंग मॉडल, 8B सक्रिय पैरामीटर प्रदर्शन दस गुना बड़े मॉडल से बेहतर","summary":"Poolside ने तीसरा प्रोग्रामिंग मॉडल Laguna S 2.1 जारी किया, जो MoE आर्किटेक्चर का उपयोग करता है, कुल पैरामीटर 118B है, प्रत्येक टोकन केवल 8B पैरामीटर सक्रिय करता है, और यह मिलियन टोकन संदर्भ विंडो और सोचने/न सोचने के दो मोड का समर्थन करता है।","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside ने Laguna S 2.1 लॉन्च किया: 118B पैरामीटर ओपन-सोर्स प्रोग्रामिंग मॉडल, 8B सक्रिय पैरामीटर प्रदर्शन दस गुना बड़े मॉडल से बेहतर - Aioga AI समाचार","description":"Poolside ने तीसरा प्रोग्रामिंग मॉडल Laguna S 2.1 जारी किया, जो MoE आर्किटेक्चर का उपयोग करता है, कुल पैरामीटर 118B है, प्रत्येक टोकन केवल 8B पैरामीटर सक्रिय करता है, और यह मिलियन ट...","url":"https://www.aioga.com/hi/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:46:06.022Z"},"it":{"title":"Poolside ha rilasciato Laguna S 2.1: modello di programmazione open source con 118 miliardi di parametri, le prestazioni di 8 miliardi di parametri superano quelle dei modelli dieci volte più grandi","summary":"Poolside ha rilasciato il terzo modello di programmazione Laguna S 2.1, che utilizza l'architettura MoE, con un totale di 118 miliardi di parametri, attivando solo 8 miliardi di parametri per token, supportando una finestra di contesto di milioni di token e modalità doppia di pensiero/non pensiero.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside ha rilasciato Laguna S 2.1: modello di programmazione open source con 118 miliardi di parametri, le prestazioni di 8 miliardi di parametri superano quelle dei modelli dieci volte più grandi - Aioga Notizie IA","description":"Poolside ha rilasciato il terzo modello di programmazione Laguna S 2.1, che utilizza l'architettura MoE, con un totale di 118 miliardi di parametri, attivando solo 8 miliardi di pa...","url":"https://www.aioga.com/it/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:46:49.356Z"},"nl":{"title":"Poolside heeft Laguna S 2.1 uitgebracht: 118B parameters open-source programmeermodel, 8B actieve parameterprestaties overtreffen modellen van tien keer de grootte","summary":"Poolside brengt het derde programmeermodel Laguna S 2.1 uit, met een MoE-architectuur, totaal 118B parameters, slechts 8B parameters geactiveerd per token, ondersteunt miljoen-token contextvenster en denk/niet-denk duale modus.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside heeft Laguna S 2.1 uitgebracht: 118B parameters open-source programmeermodel, 8B actieve parameterprestaties overtreffen modellen van tien keer de grootte - Aioga AI-nieuws","description":"Poolside brengt het derde programmeermodel Laguna S 2.1 uit, met een MoE-architectuur, totaal 118B parameters, slechts 8B parameters geactiveerd per token, ondersteunt miljoen-toke...","url":"https://www.aioga.com/nl/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:46:49.222Z"},"tr":{"title":"Poolside, Laguna S 2.1'i yayınladı: 118B parametreli açık kaynaklı programlama modeli, 8B aktif parametre performansı on kat daha büyük modeli aşıyor","summary":"Poolside, MoE mimarisi kullanan ve toplam 118 milyar parametreye sahip üçüncü programlama modeli Laguna S 2.1'i yayımladı; her token yalnızca 8 milyar parametreyi etkinleştiriyor ve milyon token’lık bağlam penceresi ile düşünme/düşünmeme çift modunu destekliyor.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside, Laguna S 2.1'i yayınladı: 118B parametreli açık kaynaklı programlama modeli, 8B aktif parametre performansı on kat daha büyük modeli aşıyor - Aioga AI Haberleri","description":"Poolside, MoE mimarisi kullanan ve toplam 118 milyar parametreye sahip üçüncü programlama modeli Laguna S 2.1'i yayımladı; her token yalnızca 8 milyar parametreyi etkinleştiriyor v...","url":"https://www.aioga.com/tr/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:47:37.090Z"},"vi":{"title":"Poolside phát hành Laguna S 2.1: mô hình lập trình nguồn mở với 118 tỷ tham số, hiệu năng tham số hoạt động 8 tỷ vượt trội so với các mô hình lớn gấp mười lần","summary":"Poolside phát hành mô hình lập trình thứ ba Laguna S 2.1, sử dụng kiến trúc MoE, tổng số tham số 118 tỷ, mỗi token chỉ kích hoạt 8 tỷ tham số, hỗ trợ cửa sổ ngữ cảnh hàng triệu token và hai chế độ tư duy/phi tư duy.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside phát hành Laguna S 2.1: mô hình lập trình nguồn mở với 118 tỷ tham số, hiệu năng tham số hoạt động 8 tỷ vượt trội so với các mô hình lớn gấp mười lần - Tin tức AI Aioga","description":"Poolside phát hành mô hình lập trình thứ ba Laguna S 2.1, sử dụng kiến trúc MoE, tổng số tham số 118 tỷ, mỗi token chỉ kích hoạt 8 tỷ tham số, hỗ trợ cửa sổ ngữ cảnh hàng triệu tok...","url":"https://www.aioga.com/vi/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:47:38.387Z"},"id":{"title":"Poolside merilis Laguna S 2.1: model pemrograman sumber terbuka dengan 118B parameter, kinerja parameter aktif 8B melampaui model besar sepuluh kali lipat","summary":"Poolside merilis model pemrograman ketiga, Laguna S 2.1, menggunakan arsitektur MoE, dengan total 118 miliar parameter, hanya mengaktifkan 8 miliar parameter per token, mendukung jendela konteks hingga jutaan token dan mode berpikir/tidak berpikir ganda.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside merilis Laguna S 2.1: model pemrograman sumber terbuka dengan 118B parameter, kinerja parameter aktif 8B melampaui model besar sepuluh kali lipat - Berita AI Aioga","description":"Poolside merilis model pemrograman ketiga, Laguna S 2.1, menggunakan arsitektur MoE, dengan total 118 miliar parameter, hanya mengaktifkan 8 miliar parameter per token, mendukung j...","url":"https://www.aioga.com/id/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:48:21.989Z"},"th":{"title":"Poolside เปิดตัว Laguna S 2.1: โมเดลการเขียนโปรแกรมเปิด 118B พารามิเตอร์, พารามิเตอร์ 8B ที่ใช้งานมีประสิทธิภาพเหนือกว่าโมเดลขนาดใหญ่สิบเท่า","summary":"Poolside ได้เปิดตัวโมเดลการเขียนโปรแกรมที่สาม Laguna S 2.1 ซึ่งใช้สถาปัตยกรรม MoE โดยมีพารามิเตอร์รวม 118B โดยเปิดใช้งานเพียง 8B พารามิเตอร์ต่อโทเค็น รองรับหน้าต่างบริบทโทเค็นนับล้านและโหมดคิด/ไม่คิดสองแบบ","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside เปิดตัว Laguna S 2.1: โมเดลการเขียนโปรแกรมเปิด 118B พารามิเตอร์, พารามิเตอร์ 8B ที่ใช้งานมีประสิทธิภาพเหนือกว่าโมเดลขนาดใหญ่สิบเท่า - ข่าว AI Aioga","description":"Poolside ได้เปิดตัวโมเดลการเขียนโปรแกรมที่สาม Laguna S 2.1 ซึ่งใช้สถาปัตยกรรม MoE โดยมีพารามิเตอร์รวม 118B โดยเปิดใช้งานเพียง 8B พารามิเตอร์ต่อโทเค็น รองรับหน้าต่างบริบทโทเค็นนับล้...","url":"https://www.aioga.com/th/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:48:36.047Z"},"pl":{"title":"Poolside wydaje Laguna S 2.1: 118B parametrów otwartoźródłowy model programowania, 8B aktywnych parametrów przewyższa wydajność dziesięciokrotnie większych modeli","summary":"Poolside wydał trzeci model programistyczny Laguna S 2.1, wykorzystujący architekturę MoE, o łącznej liczbie parametrów 118 miliardów, przy czym na każdy token aktywowanych jest tylko 8 miliardów parametrów, obsługujący milionowy kontekst tokenów oraz tryb myślenia/niemyslenia.","category":"模型更新","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Poolside wydaje Laguna S 2.1: 118B parametrów otwartoźródłowy model programowania, 8B aktywnych parametrów przewyższa wydajność dziesięciokrotnie większych modeli - Aioga Wiadomości AI","description":"Poolside wydał trzeci model programistyczny Laguna S 2.1, wykorzystujący architekturę MoE, o łącznej liczbie parametrów 118 miliardów, przy czym na każdy token aktywowanych jest ty...","url":"https://www.aioga.com/pl/news/cmrxihnkj00e9ro4ju90yp4sr/","contentTranslated":true,"sourceHash":"879a007d13d60081","translatedAt":"2026-07-27T00:49:20.222Z"}}}}