搜索“Agent”

18 evaluations (higher is better) GPT‑6.1 Sol scores higher than GPT‑6 Sol in five of six categories; the regression on the gore evaluation is not statistically significant. 3.1.3 Agentic Safe Completions Table 3: Agentic Safe Completions (higher is better) We find that GPT-6.1 Sol is generally better than GPT-6 Sol at recognizing potential harms in agentic requests and avoiding taking harmful actions. 3.2 Vision Table 4: Image input evaluations, with metric safe completion (higher is better) Ac

匹配字段正文

otein engineering) were asked to transcribe biological protocols they have personally used in the lab. Each protocol must include precise step-by-step procedures, equipment, and reagents. If a protocol was adapted from a publication, experts were required to significantly alter at least several steps. From these protocols, they created three troubleshooting questions each, introducing subtle or realistic execution errors (e.g., improper homogenization technique) and describing the resulting fail

匹配字段正文

a Model API and Muse Code. Muse Code installs with curl dev.meta.ai/install.sh | bash (@alexandr wang). Positioning. Meta pitches the models as SOTA for “personal and professional agentic work.” @alexandr wang named three headline features: Active Memory: memory across days, weeks and months. Active Focus: working through blockers to finish tasks. Domain expertise: SOTA in finance and legal, “strong in health.” Training story. Wang says Meta continued scaling both pre-training and RL since Muse

匹配字段正文

; Reply 0 Meta Platforms announced Oct. 13 that it has unveiled its latest frontier artificial intelligence (AI) models, the Muse Wonder series, featuring major improvements in AI agents' ability to independently handle complex tasks. The series comprises two versions, Muse Wonder and Muse Wonder Full. The models are differentiated by their availability to external developers and their use within Meta's proprietary work environment. Both are available only through paid subscriptions priced at $5

匹配字段正文

[[ Meta新旗舰模型Muse Wonder在长程编程等Agent测试中超越Claude Opus 5.5，综合智能指数逼近第一梯队，但推理成本大涨183%，且3D生成实测表现不佳，暴露视觉空间推理短板。 要点 1. 旗舰跑分挤进第一梯队，长程编程领先 Wonder Full综合智能指数56分，距Claude Opus 5.5仅差2分；DeepSWE 1.1长程软件工程测试中取得79.0分，超过Opus 5.5的74.2分和GPT-6 Astra的74.1分，在10项测试中有6项跻身前三。 2. 换用新预训练底座，针对Agent能力优化 Wonder内部代号Watermelon，采用独立更大规模预训练并引入Active Memory与Active Focus机制，从底层增强模型的长期记忆、持续执行与任务泛化能力，而非像Spark系列那样仅依靠后续强化学习迭代。 3. 推理成本大幅上涨，颠覆性定价策略引发质疑 Wonder任务平均成本达4.52美元，较上代Spark 1.3上涨183%，比GPT-6 Astra高39%；官方虽小幅降低按Token计费价格，但Full版本的订阅额度消耗异

匹配字段正文

r Full新模型至少在部分基准测试上，已能与GPT-6 Astra和Claude Fable 5.1掰手腕。 第三方机构Artificial Analysis的测试也显示，Wonder Full较上一代脱胎换骨，在多项智能体和专业工作评测中跻身前三甲。 花大价钱攒起来的队伍，让Meta重新有了跟领先模型叫板的底气，也让Meta Muse的Personal Agent之路能走得更顺畅些。 Meta此次发布，延续了一年多来对AI业务的重整。 2025年4月，Llama 4马失前蹄，扎克伯格随即加大人才和算力投入，招揽Scale AI创始人Alexandr Wang，重组研究团队。今年4月，重组后的团队推出Muse Spark，但与竞争对手相比仍差着一口气。 按照Meta的说法，过去14个月里，团队推倒重来，从底层重建技术体系，并持续扩大预训练和强化学习的规模。此次Wonder系列在吉瓦级Prometheus数据中心完成训练，成为这轮豪赌交出的最新筹码。 Muse Wonder围绕两个方向动刀，分别是Active Memory和Active Focus。 Active Memory负责跨对

匹配字段正文

The Muse Wonder series achieves state-of-the-art results across key benchmarks for personal and professional agents, memory, and coding. Muse Wonder and Muse Wonder Full apply specialized knowledge to problems people encounter at work and in daily life, from understanding a lease to making sense of financial or health information. Our evaluations include tasks in law, finance, and health that test this ability.

匹配字段正文

We’re excited to introduce the Muse Wonder series with Muse Wonder and Muse Wonder Full, our most capable models yet. Both are frontier models, built to deliver state-of-the-art performance for personal and professional agentic tasks.

匹配字段正文

same update cut Clef-flash pricing below TypeSafe’s Jev and made hosted Clef roughly 2x faster. For developers, this removes the transcription and captioning steps that multimodal agent pipelines usually need. TL;DR Size: 30B total, 3B active parameters (MoE); 64,000-token context on Workers AI; Runs on: Workers AI as @cf/cloudflare/clef-omni; self-hosting tested on a single H200, about 64 GB GPU memory in bfloat16; Performance: Close to Clef on routing and intent tasks, behind it on most reason

匹配字段正文

d variants, hardware not disclosed; Performance: Vendor-reported scores are near ceiling on cyber benchmarks and strong on coding; Best: 100% on Cybench (39/39 tasks, unrestricted agent execution); Worst: 76.5% on SWE-bench Pro V2, its lowest published score; Bottom line (best): Top-tier cyber scores at $3.00 / $7.50 per 1M tokens; Bottom line (worst): Every number is self-reported, and CVE-Bench used only 24 evaluable tasks What is OrcaCyber Zero 1.5? OrcaCyber Zero 1.5 is a frontier cybersecur

匹配字段正文

owcase what Inkling can do and how it measures up on important qualities such as trustworthiness and safety. Generalist model Inkling is designed to be broad. We trained it across agentic, reasoning, coding, instruction-following, factuality, vision, and audio tasks, rather than narrowly optimizing for one domain. That breadth matters for customization and real-world use: different users need models that can adapt to very different workflows, not just excel on benchmarks. Agentic coding and tool

匹配字段正文

and well-rounded performance across a range of benchmarks. Inkling-Small Inkling Compared to Inkling, Inkling-Small achieves comparable performance with much less compute. Across agentic tool use (Terminal-Bench 2.1), reasoning (HLE text-only, no tools), and instruction following (IFBench) benchmarks, Inkling-Small is more efficient than Inkling and competitive with other models in its weight class. Furthermore, its variable thinking effort lets users easily adapt it to target their use case, b

匹配字段正文

ducing Beam, Reflection’s first open-weight model. Beam is a sparse Mixture-of-Experts model with 501 billion total parameters, 23 billion active, built for coding, reasoning, and agentic workloads. Beam’s capabilities come from major investments in both pretraining and reinforcement learning (RL). We pretrained the model on 23.8 trillion diverse, curated, high-quality tokens from the web and proprietary licensed datasets, matching or outperforming available similar-sized open base models. In pa

匹配字段正文

on-1 is built for fast, structured decisions. According to Microsoft , it handles classifications, evaluations, and routing decisions and has "the potential" to "guide and control agents through complex environments." Decision-1 is based on Qwen3.5-9B. Microsoft says it's the most accurate model tested across 36 benchmarks covering nearly 150,000 questions. On speed, it's 2.5 times faster than the runner-up, H2O-Lightning-4B. Cloudflare's open-source Clef models, which are also based on Qwen, we

匹配字段正文

c.twitter.com/M3fAOD90r6 — Satya Nadella (@satyanadella) October 9, 2026 Xbox Research used it to sort more than 10,000 pieces of player feedback, the Copilot team graded chat and agent responses with it, on-call engineers used it to pull context during live incidents, and Microsoft Discovery used it to score experiments before an agent replans. Microsoft’s numbers show it’s more than 14 times as fast as GPT-6 Sol at a fraction of the cost for Xbox, and 46 times more consistent in Discovery. But

匹配字段正文

unlocks advanced photo-editing tools, the all-new Image Playground, and more; AFM 3 Cloud Pro, our most capable server-based model, which powers our most demanding use cases, like agentic tool use and complex reasoning AFM 3 Core, AFM 3 Core Advanced, and AFM 3 Cloud, along with ADM 3 Cloud, are all purpose-built for Apple silicon. For AFM 3 Cloud Pro, we worked with Google and NVIDIA to extend Private Cloud Compute to NVIDIA GPUs in Google Cloud, while maintaining the same guarantees to protect

匹配字段正文

Nace.AI has open-sourced Drex 1.5, a 9B decision model for agents and backend workflows. The Drex 1.5 decision model does not write text. It reads a state and typed questions, then returns a probability for every option. Nace reports 58.08 on the public Decision Index 0.3.1, the top score under 10B parameters. Weights are on Hugging Face, and a hosted version is live on OpenRouter. TL;DR Size: 8.95B parameters (dense), bf16 weights about 18 GB. Context is 16,384 tokens by default, up to 131,072;

匹配字段正文

Microsoft has released Microsoft-Decision-1, a decision model for routing, classification, verification and agent control. Microsoft-Decision-1 is a decision-scoring model that returns a calibrated probability for each fixed answer option instead of generated text. It is post-trained from Alibaba’s Qwen3.5-9B and available now in Microsoft Foundry and OpenRouter. . TL;DR Size: Built on Qwen3.5-9B; exact parameter count not disclosed. 32,768-token context window; Runs on: Hosted API only (Microso

匹配字段正文

编码任务，Gemini 4 Pro拿下88.7%，超过Astra的86.9%； GDPval-AA v2真实知识工作任务，它是唯一突破2000 Elo大关的模型，达到2064； Terminal-bench 2.1终端编码能力95.3%，全场最高；OSWorld-2.0计算机操作测试86.8%，同样排名第一。 在Terminal-Bench 4.0，专门考察Agent连续多步执行能力的测试，4 Pro与自家Flash的差距从上一代测试中的3.2%，直接拉大到13.9%。 换句话说，任务越复杂、链路越长、步骤越多，4 Pro的优势就越明显。 这恰恰是当前所有AI Agent最致命的短板——在需要持续串联多步动作、维护长上下文、反复读取状态并重新规划时，绝大多数模型都会出现严重的性能衰减。 而Gemini 4 Pro，似乎在这一点上找到了突破口。 价格方面更是让人意外。每百万Token输入仅2.25美元、输出11.25美元，在顶级旗舰中堪称“价格屠夫”。 黑进三家公司 内部Gemini干的 就在所有人还在为“机械蝴蝶”惊叹时，一个更令人脊背发凉的消息传了出来。 Google内部模型，可能是

匹配字段正文

hosting needs separate permission; No Turbo-specific benchmark exists yet; base Qwen-Image-2.1 scores 60.28 on Qwen-Image-Bench Asif Razzaq Asif RazzaqGoogle Cloud Launches Gemini Agent, One Universal Agent for Enterprise Work; Asif RazzaqJetBrains Releases Mellum2.1: A 12B MoE Open Model for Coding Agents; Asif RazzaqPerplexity AI Releases pplx-embed-v2-late: A 0.6B Edge Model and a 9B Model Scoring 92.4% on MADQA; Asif RazzaqWhat Happens When a Trusted Model Repo Changes? Unsloth Studio Re-Che

匹配字段正文