AI & Machine Learning

See who said yes to your team’s AI agents

Cloud computers for any AI agent, customized your way

Small, fast models for reranking, embeddings and decisions

A team of AI agents for video: produce, edit, clip, post

Give AI agents their own screen on your Mac

SSH client with an AI agent and a desktop for every server

Point at your app and your AI agent changes the code

New AI agent, 1,650+ apps, Skills and smart model routing

Cloud browser for AI agents and tough sites

Run AI agents with Slack, Supabase and 50+ other services

Hold ⌥Space, flick, and an AI agent does the dev chore

Hand your financial admin to your AI agent

Agent decisions depend on empirical relationships that deployment observations reveal and task descriptions alone may leave unresolved. We introduce Decoupled Predictive Decision Making (DPDM), an architecture that formulates heterogeneous agent decisions as conditional prediction with a shared pretrained learner. Language models provide semantic representations of current and historical records, while a general-purpose conditional predictor predicts task-defined targets from the represented evidence. Pretraining supplies reusable inference capability; deployment evidence instantiates problem-specific predictions without updating either model's neural parameters. Tool judgments, candidate preferences, and reasoning assessments share an evidence-and-query interface, with predictions mapped to choices by a decision rule. This architecture establishes complementary sources of agent capability: semantic knowledge acquired through language pretraining and empirical inference acquired through structured-data pretraining. A single frozen LimiX-2 predictor serves all five task categories. With Qwen3.5-27B as the semantic encoder, DPDM raises mean accuracy across 50 benchmarks from 78.80% with few-shot prompting to 84.92% under their respective evaluated history budgets, and reaches 82.33% with identical histories and queries. On ToolPRMBench's BFCL and ToolSandbox tasks, absolute accuracy gains are 34.26% and 28.87%, respectively, at the methods' respective evaluated history budgets. Accuracy improves as history grows. With identical imbalanced histories, mean balanced accuracy improves by an absolute 14.51% across five reasoning-step assessment tasks and four encoders. DPDM makes learning from experience an independently improvable foundation-model capability, creating a shared route for advances in conditional prediction to improve decisions across agent tasks.

Models for physical AI tasks, such as robotics and autonomous driving, need not reason solely in language. They can use visual plans--points, bounding boxes, and trajectories--to represent spatial goals and anticipate motion. However, autoregressive vision-language models generate these plans one coordinate token at a time, making richer predictions increasingly costly. We introduce VFlash, a method for fast, accurate visual planning through Flash Primitive Decoding. VFlash exploits the known structure of visual primitives to predict their coordinates jointly, using a lightweight drafter that generates an entire primitive in one forward pass. The layout adapts to the primitive's type, dimensionality, and length, supporting compact boxes and long, variable-length trajectories. Crucially, spatial accuracy need not require reproducing an autoregressive model's exact token sequence. VFlash therefore trains on annotated primitives and directly accepts structurally valid predictions without target-token agreement. Across the evaluated 2D and 3D robot trajectory tasks, VFlash consistently reduces prediction errors relative to autoregressive decoding. Preliminary single-rollout measurements indicate approximately 25x faster generation for 3D point motion forecasting. A world action model conditioned on VFlash plans achieves 70.2% average success across four MolmoSpaces task suites, compared with 62.6% for the baseline.

Harness evolution from a common base exhibits Locality, Complementarity, and Nonstationarity: variants repair different failures, and their utility changes as the policy learns. Rather than seeking a universally superior harness, we introduce Harness-S to selectively rescue failures of a base execution system through sibling harnesses and transfer the resulting capabilities into a single policy that remains effective under the original harness. Directional Harness Evolution develops specialized sibling harnesses along distinct repair directions, forming a reusable Sibling Harness Pool that remains fixed during policy training. Complementary Harness Rescue Training invokes this pool only when all initial attempts under the base harness fail. The same policy generates fresh rollouts under selected auxiliary harnesses, and these trajectories replace the failed group for reinforcement learning. An Adaptive Harness Router selects the Top-k variants using Harness Behavior Profiles and voting across multiple LLMs. Policy-aware profile re-anchoring refreshes this evidence as the policy learns. Across AppWorld, SWE-Gym, and DBBench, the trained policy substantially outperforms policy-only training when returned to the original base harness. Behavioral analyses and held-out cases support the transfer of harness-elicited behaviors into the policy itself. Reusing the router for Rank-1 inference provides additional gains at the same executor rollout count.

Benchmark compression reduces the often prohibitive time and cost of model evaluation by evaluating models on a small benchmark subset that closely matches full-benchmark results. To quantify estimation error, we derive an accurate analytical approximation to the error distribution when estimating full-benchmark scores from a subset, showing that most subsets yield small errors, while a few produce much larger errors, resulting in a long-tailed distribution. This means random sampling alone can achieve surprisingly strong performance with high probability, but remains unstable due to rare large errors. We further show that this instability is inherent to modern foundation models and persists even as more models are evaluated. Based on these findings, we introduce BestRand, a stable and low-cost benchmark compression method that preserves the efficiency of random sampling while substantially reducing its high-error tail. We validate BestRand on 70+ text, multimodal, and agent benchmarks, and further extend it to VLA evaluation. Compared with competitive methods, BestRand reduces high-error risk by 50.2% and estimation error by 16.4% on average, while requiring only 51.3% of the construction cost. It further reduces OOD estimation error by 21.8% on average across score-range, model-family, and temporal shifts. Together, our theory provides practical guidance for understanding and predicting the accuracy and reliability of benchmark compression, while BestRand enables stable and accurate compression at low cost. The code and data are publicly available onhttps://github.com/George-Ling3/BestRand.

Reliable robotic manipulation requires both semantic understanding and precise control in three-dimensional space. However, vision-language-action (VLA) foundations largely inherit visual representations from 2D vision-language pre-training, while spatial supervision is often introduced only during task-specific adaptation. Developing spatial understanding as a transferable foundation-model capability therefore calls for learning from broader robot experience. We present SLA, a Spatial-Language-Action Model that scales implicit spatial alignment to continual pre-training and maintains it during downstream post-training. Initialized from a pretrained pi05 policy, SLA jointly optimizes action prediction and alignment between intermediate visual embeddings and spatial representations from a frozen 3D perception model. We curate and unify heterogeneous robot datasets to support spatial learning across diverse embodiments, tasks, and scenes. To accommodate this diversity, we introduce feature-dependent gates that adaptively aggregate visual embeddings from multiple VLA layers for spatial supervision. The teacher further exploits partially available depth and camera calibration alongside RGB observations, enabling effective use of heterogeneous geometric annotations. Probabilistic state dropout encourages reliance on visual cues and mitigates overfitting to demonstration trajectories. All auxiliary components are used only during training, allowing direct reuse of pretrained VLA weights and preserving the original inference architecture.

We present WeLM-Image, a unified image foundation model that integrates multimodal understanding into text-to-image generation, instruction-guided editing, and multi-reference synthesis. Designed for single-GPU deployment in WeChat's large-scale consumer applications, the model processes text, reference images, and target latents as one interleaved sequence. A Mixture-of-Transformers (MoT) backbone combines shared attention with independent understanding and generation parameters to limit optimization interference. Equipping each path with Mixture-of-Experts (MoE) modules provides sparse capacity, with 31.7B stored parameters and approximately 5.4B generation parameters activated per target latent token. A generation-oriented data engine combines a knowledge-grounded visual vocabulary with large-scale visual clustering to balance semantic coverage and editing tasks, while training-time evaluation guides iterative data refinement. Bilingual annotations and verified editing supervision support progressive training up to native 1.5K resolution. We optimize generation and editing separately with rubric- and checklist-based rewards, then consolidate the specialized policies through noise-aware on-policy distillation. In the reported overall comparisons, the 1536-resolution model ranks first among the evaluated open-source models on PRISM-Bench, Qwen-Image-Bench, and UnicBench. MeanFlow-based few-step distillation yields WeLM-Image-Turbo, which generates a 1024times1024 image in 4.98\,s on a single NVIDIA H20 GPU using eight guidance-free steps.

Strong single-turn editing performance does not guarantee reliable multi-turn interaction: as edited outputs are repeatedly reused, small unintended changes can accumulate into severe visual drift. This long-term drift is often mitigated through training on multi-turn data or cross-turn rollouts, potentially altering pretrained priors, or through specialized inference-time mechanisms that are difficult to scale to foundation models. We introduce TailSteer, a final-stage training framework that improves multi-turn robustness while retaining the generative and editing capabilities acquired through earlier training. TailSteer follows a simple principle: keep early denoising dynamics close to those of the base model and concentrate preservation corrections near the trajectory tail, where fine-grained deviations can be suppressed. These corrections are learned from the editor's own on-policy denoising rollouts, using unchanged source content as a preservation reference without requiring curated ground-truth editing pairs. Experiments across the FLUX, SenseNova, and Qwen families show that TailSteer substantially reduces cumulative drift over 15-20 editing turns while largely preserving base-model performance on extensive text-to-image and single-image editing benchmarks. These results demonstrate that long-horizon editing robustness can be unlocked in foundation models after TailSteer, rather than by relearning their existing capabilities.

Autoregressive approaches with large language models, such as long chain-of-thought reasoning, have become the default for new AI applications because they reach higher accuracy on a wide range of tasks. In production, however, they give inconsistent answers, require parsing of outputs, and add cost and latency. Decision models address this: given a question and a set of options supplied by the user, they return a probability distribution over the options without autoregressive decoding. We present Drex, a family of four open decision models with 8B (DLM), 9B (v1.5), 35B with 3B active (v2.0), and 27B (Max) parameters. Each request consists of a state (a document, record, or agent context, optionally with images) and questions about it, each asking for a choice among options, a yes/no judgment, or a score on an ordered scale. Training combines supervised fine-tuning with reinforcement learning from adjusted feedback (RLAF), plus vision tuning for Max. Ablations show that RLAF improves accuracy and that consistency losses improve robustness to option order. On Decision Index 0.4.1, Max sets a new state of the art among the decision models we compare (67.52 on the public suite vs. 60.70 for Jev 1.14.0). On Decision Bench v4, it is statistically tied with d3-turbo and GPT-6 Sol and trails Claude Opus 5.5 by 1.6 points, and it has the highest accuracy among decision models on DecideBench. Max answers a typical request in 183 ms, against about 2.0 s for GPT-6 Sol and 3.5 s for Claude Opus 5.5, and its outputs are stable under option reordering. All four models are open source: https://github.com/nace-ai/drex-decision-models

Streaming video understanding requires models to repeatedly decide whether to respond or continue observing. Recently, Group Relative Policy Optimization (GRPO) has been applied to this setting, with existing approaches emphasizing reward design while retaining a single trajectory-level advantage across turns. Such shared credit can reinforce redundant replies after earlier successes and discourage useful follow-ups after earlier failures. We propose TurnWise, a turn-level GRPO framework that aligns learning with individual streaming decisions. TurnWise addresses the mismatch between trajectory-level feedback and turn-level interaction, promoting informative responses and reducing redundancy. Specifically, turn-level credit assignment uses silence-completed prefix scoring to construct group-relative advantages from each decision's remaining return, separating its learning signal from earlier rewards. Turn-mean policy optimization averages token losses within each turn and then across turns, giving response and waiting decisions equal weight regardless of output length. These complementary designs improve streaming policy learning without a learned critic or additional rollouts. Extensive experiments show our method outperforms prior RL approaches for streaming video understanding.

Vision-language-action models are typically built on a simple premise: perception is first compressed by visual encoders, and action reasoning begins only afterwards. Yet manipulation often hinges on precisely the visual signals most vulnerable to this abstraction, e.g., local geometry, boundaries, depth, and subtle state changes. This raises a fundamental question: should action policies inherit a fixed visual endpoint, or directly access evolving visual computation? We introduce NEO-VLA, a structurally encoder-free vision-language-action framework. Images are tokenized by lightweight patch convolutional layers and evolve jointly with language throughout the backbone, allowing a flow-matching action expert to draw from instruction-conditioned visual representations at different depths. This simple change reveals a consistent pattern. Native architectures outperform encoder-based counterparts under controlled comparisons; full fine-tuning reduces but does not erase the gap, and additional scale alone cannot recover it. More strikingly, layer-wise probing shows that the action-critical signals often peak well before the final visual representation. Across diverse manipulation settings and distribution shifts, NEO-VLA consequently achieves stronger and more robust control. Together, these results suggest that embodied control benefits from access to visual representations across the computational hierarchy, rather than a fixed encoder output.

AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research quality alongside correctness. We introduce FrontierPhysics, an end-to-end benchmark with 56 tasks made by 42 physicists based on their own research, covering nine domains across theoretical, computational, and experimental physics. Each task requires agents to review the literature, develop a research plan, conduct research and experiments, and produce structured outputs and scientific manuscripts. Combining deterministic verifiers with expert-authored rubrics, FrontierPhysics evaluates whether agents' research is correct and publishable. Across 22 models in 26 configurations, no configuration passes more than 46 of its 168 rollouts. Manual review of agent trajectories shows that frontier agents' main weaknesses are insufficiently rigorous reasoning and limited expert intuition, rather than gaps in physics knowledge or execution errors.

A frontier AI model is trained in an AI data center, on tens of thousands of accelerators joined by a dedicated network, and every training step ends with all of them exchanging data at once. Between racks, that network is optical, and each of its four layers, from lasers to switches, runs a fixed or after-the-fact policy. Lasers sharing an enclosure, cooling, and power can fail together. At a fixed spare ratio, correlated failure provably caps a pool's availability however large it grows. In this paper, we present a comprehensive predictive control plane framework, called Halley, that reads each resource's next state and acts before the failure occurs. It has four components, each carrying a formal guarantee: LightReserve that sizes the laser pool to provide a coverage bound, FoldFit that schedules the switch within 11/9 of the optimal, ThermaLead that holds a microring with one-step-ahead control, and TriageGuard that protects the highest-risk modules first. Halley gives each layer a predictive controller that defers to the deployed policy when unsure. Composed from simulations at that point, effective goodput, the share of ideal throughput kept, rises from 65.2\% to 76.5\% if a window that exhausts its spares is lost. If only the dark channels' bandwidth is lost, it reaches 67.4\%. Per megawatt, that is up to 1.03--1.17 times the training; for the same training, up to 3.3--14.7\% less data center capacity. Halley's fallback circuits are formally verified and are shown to achieve close timing on a commercial FPGA, and pass a 130~nm signoff.

Natalia Kniazhevich / Bloomberg : The Nasdaq 100 drops 6%, losing a record $2.3T in value, and the Philadelphia chip index plunges 12%, on concerns over the AI trade after Democrats won Congress —  Technology stocks plunged the most since 2020 on Thursday as investors exited the sector on concerns over the artificial intelligence trade …

Jay Peters / The Verge : Meta rolls out voice mode and expressive avatars in the Muse app, starting in the US, letting users have real-time voice conversations with their Muse AI agent —  Muse AI agent now has voices and ‘expressive’ avatars that animate while chatting with you.

Reuters : Sources: about 600,000 of Nvidia's H200 chips are stored in Hong Kong awaiting shipment to mainland China, as Beijing withholds approval for large-scale imports —  About 600,000 of Nvidia's advanced artificial intelligence chips are stored in Hong Kong awaiting shipment to mainland China …

Mark Gurman / Bloomberg : Sources: Apple cut some engineers' daily AI spending cap to $200 from $300 and plans cuts to as little as $100, as cost cutting stokes staff fears of layoffs —  Apple Inc. is reducing employee budgets for artificial intelligence usage as part of a cost-cutting effort, even as the company seeks …

Rashi Shrivastava / Forbes : Traversal, which provides AI agents for site reliability engineering, raised $120M co-led by Kleiner Perkins, Redpoint, and Khosla at a $1B valuation —  Traversal has raised $120 million in funding to help companies like American Express and Pepsi troubleshoot site outages.  —  for The Prompt

Tobias Mann / The Register : Revolut says it cut AI coding inference costs by ~90% in less than a year by making self-hosted GLM models its default over commercial tools like Claude Code —  Revolut says that in less than a year it's managed to slash its AI coding inference costs by roughly 90 percent by making self-hosted models …

Rikiya Takehi : Researchers say LLMs can be good at natively reading and writing neural network weights; GPT-6 Astra edited Qwen3.5-0.8B's weights to make it better at math —  TL;DR … If you were asked to handcraft a neural network that plays Go, what would you do?  If you were handed raw neural network weights …

Andrew E. Freedman / Tom's Hardware : When talking about local inference, Sam Altman says supply-constrained DRAM is better used in data centers than sitting in PCs and “being used 5% of the time” —  With data centers full of RAM, your PC is apparently the problem. … OpenAI CEO Sam Altman has data centers full of memory …

Raga Justin / Bloomberg Government : NY Gov. Kathy Hochul signs bills banning AI chatbots from using “unsafe features” for minors and requiring security features in internet-connected baby monitors —  New York Gov. Kathy Hochul signed two bills further curbing artificial intelligence risks for children …

Bloomberg : Sources: AI robotics startup Physical Intelligence is in talks to raise $1B+ led by Avenir and Lux Capital at a ~$20B valuation, up from $11.5B earlier in 2026 —  Physical Intelligence, a startup developing artificial intelligence software to power robots for real-world tasks …

Annie Palmer / CNBC : Memo: Amazon says measuring AI adoption solely by token usage or code generation is “deeply flawed,” as it discourages staff from tokenmaxxing —  Amazon laid out its AI usage guidelines for software engineers in a memo to staffers, and discouraged employees from turning to …

Seema Mody / CNBC : Oracle adds ChatGPT and Codex to its app stack and expands Oracle Cloud Infrastructure lineup with Google TPUs, OpenAI's Jalapeño chip, and Cerebras hardware —  Oracle on Tuesday announced an expanded collaboration with OpenAI that brings the artificial intelligence lab's ChatGPT …

Linus Torvalds 在内核邮件列表上宣布释出 Linux 7.4-rc1，关闭合并窗口，正式版本预计在圣诞节前发布。Linux 7.4 共包含了 18,591 个非合并（non-merge）变更集，这一数字创下了纪录；有 2,678 名开发者为 v7.4 贡献了代码，其中 625 人为首次。Torvalds 表示他乐见 AI 相关工作在合并窗口期间大量涌入，Linux 7.4 合并窗口合并的补丁规模高于通常的版本，显示 AI 修复逐渐融入常规开发流程，他认为这是一种更健康的模式。Linux 7.4 的新特性包括：Radeon iGPU 支持 AMD PerfOpt 帮助增强 AI/LLM 性能，进一步完善对 AMD Ryzen Zen 6 架构处理器的支持，默认启用 AMDGPU HDMI 2.1，英特尔新 Nova Lake 笔记本电脑支持 IPU8，更多高通 Snapdragon X2 笔记本电脑支持，EXT4 可选支持缓冲 IOma，等等。

在人类演化史的大部分时间里，地球同时存在多个人种。以过去 10 万年为例，尼安德特人曾活跃在欧洲和西亚，丹尼索瓦人生活在亚洲，形似霍比特人的弗洛勒斯人（Homo floresiensis）居住在印尼的弗洛勒斯岛，爪哇岛上最后的直立人刚刚灭绝。现代人类/智人只是其中之一。我们自己可能会认为如今单一的的人类世界是常态，研究远古历史的科学家则认为如今才是奇怪的时代，为什么现代人类成为最后幸存的一支人种？其他人种都经历了什么？在一种叙事中，人类认为其近亲过于原始而理所当然被淘汰，认为现代人类有着更大的大脑、熟练的工具使用能力以及复杂的社会结构而注定要统治地球，我们是最优秀的。但越来越多的证据显示，我们与近亲的智力差距其实很小。尼安德特人由于生存年代较近而被研究的最为充分。那么尼安德特人究竟发生了什么？一种回答是：他们仍然活着，尼安德特人与现代人类融合，他们的 DNA 留在我们的 DNA 里。丹尼索瓦人的情况类似。当然尼安德特人自己可能不会这么认为，他们作为独立的人种已经消失了。那么为什么如今只剩下现代人类？我们的成功有多少源于自身独特的能力？又有多少仅仅归于运气？澳大利亚研究组织 Australian Research Council’s Centre of Excellence for Transforming Human Origins Research 主任 Michael Petraglia 指出，我们必须认识到，智人在地球上存在的时间其实并不长，我们的生存并非理所当然。智人最终能否被视为成功的物种仍未成定论。直立人曾在地球上生存了逾一百万年，尼安德特人则存在了四十万年。尽管科学界对于智人的起源仍存争议，但许多学者认为，智人作为一个物种至今仅存在了约三十万年。