AI News

Build cloud agent coworkers that take on real, multi-step work across the tools you already use. Describe the outcome you want, and Skydive creates a working agent in minutes. No code, no prompt engineering, no workflow wiring required. Skydive agents live in your stack, execute repeatable work, and get sharper over time.

Hermes Agent can now seamlessly browse as you.

Turn on real-profile browsing and your agent acts with your logins, from a managed copy of your existing Chrome profile.

🚀 New blog: MiniMax-H3 on 8×H200: 1.95× Lossless, Up to 6.24× at 0.76–0.91 SSIM

Same prompts, seeds, resolution, FPS, and 50 denoising steps held fixed across six workloads:
- Dense lossless path: 1.85–1.95× over Diffusers, no approximation
- Step reuse + sparse attention:

Evolution Strategies (ES) have recently emerged as a memory-efficient post-training paradigm for LLM reasoning. However, the optimization behavior of ES remains understudied, making it hard to define its advantage scope compared to mainstream post-training paradigms (e.g., Group Relative Policy Optimization (GRPO)). By systematically investigating ES dynamics and mechanisms, this paper first identifies a performance advantage of ES over GRPO, theoretically and empirically showing that ES can lead to broader reasoning coverage, thereby better exploiting the reasoning capabilities of pretrained LLMs. Theoretically, we show that verifier-projected Jensen-Shannon diversity across the ES population is helpful to higher Pass@K performances. Empirically, unlike GRPO, which exhibits entropy collapse, ES improves Pass@1 while attaining higher Pass@K than GRPO. We further develop a sequential GRPO-ES training strategy that combines GRPO's strength in Pass@1 with ES's gains in Pass@K. Second, we find that despite substantial whole-model parameter drift, the task-performance gains of ES are only contributed to a sparse subset of larger-magnitude updates. This functional sparsity suggests that large parameter movement need not imply widespread functional change, and held-out evaluations further show that it does not necessarily lead to catastrophic forgetting. Finally, we study how hyperparameter design affects the effectiveness of ES, demonstrating that ES requires a smaller population size in a larger LLM. These findings position ES as a distinct reasoning post-training paradigm rather than a less effective, memory-efficient alternative to GRPO.

Video carries the temporal structure of the physical world, yet learning representations from it has remained computationally expensive: prevailing self-supervised methods either prevent representation collapse through architectural asymmetries, coupling an exponential-moving-average target encoder, a stop-gradient, and a capacity-limited predictor, or circumvent it by reconstructing masked content in pixel space. We introduce LeVJEPA, the first video encoder trained under LeJEPA's collapse-free objective, which dispenses with both. A single encoder is trained with an invariance loss over global and local views of a clip, regularized by SIGReg, which excludes collapse with a provable guarantee. The architecture reduces to an encoder and a projector, and the objective to a single hyperparameter. This formulation admits two properties. First, the cost of pretraining is governed by the number of tokens the encoder observes; uniform random token dropping renders this number small while simultaneously improving downstream accuracy. At matched epochs on identical data, LeVJEPA matches or surpasses V-JEPA 2 across ViT-S/B/L at 5.6 to 20.8x less pretraining compute, and at matched total FLOPs it exceeds the strongest video baseline by 7.6 points on ImageNet-1K while remaining competitive on motion-centric benchmarks. Second, since no asymmetry between branches is required, the encoder can be trained with block-causal attention at no measurable accuracy cost: temporal ordering becomes a property of the encoder itself. Against a compute-matched DINOv2 trained on frames of the same videos, LeVJEPA approaches the image-pretrained encoder on appearance-centric evaluation while nearly doubling its motion-centric accuracy. These results indicate that, once its computational overhead is removed, video becomes a viable and in several respects preferable substrate for general-purpose visual pretraining.

OpenAI is preparing a second $400 million venture fund, far larger than its original $175 million vehicle, according to reports.

Perceptron offers an AI model that it says can help machines navigate the world while also providing in-depth visual intelligence.

fx is Vercel's tiny, open-source coding agent built to get out of your way. Written in Zig and shipped as a ~6MB native binary, it starts almost instantly while keeping memory and context overhead low. Use it with local or cloud models, extend it with skills, plugins and MCPs, or embed it into your own agent infrastructure. Small by design, so the model gets more room to work.

Bedrock Robotics deploys autonomous excavators to active job sites, tackling construction labor shortages with AI-powered heavy machinery.

Amazon Web Services (AWS), an Amazon.com, Inc. company (NASDAQ: AMZN), and NVIDIA (NASDAQ: NVDA) today announced a major expansion of their strategic collaboration to meet surging global demand for AI infrastructure as demand continues to accelerate.

Hugging Face unveiled its $399 Microduck waddling robot, expanding a hardware lineup that already includes the Reachy Mini and a growing robotics

SoftBank is in talks to buy a majority stake in OpenAI-backed humanoid robotics startup 1X Technologies in a deal that would value the company at roughly $6 billion, according to an exclusive report from The Information. The talks are still underway, and the terms could change, the report said. A deal would give SoftBank control

Nous Portal has a reworked model catalog page with everything in one place: active free models, promos, and a handy sortable list of all other available models.

https://t.co/nQtnF2YDpo

Recent prominent post-training methods, such as Reinforcement Learning (RL) and On-Policy Self-Distillation (OPSD), have driven rapid progress in mathematical reasoning for large language models, yet their reliance on ground-truth labels precludes test-time training (TTT). Replacing ground truth with majority-vote pseudo-labels is a natural alternative, yet it is fragile: an incorrect vote corrupts the teacher and misleads every token. We observe that this failure mode is asymmetric: rollouts that disagree with the pseudo-label are typically wrong regardless of whether the vote itself is correct. Building on this observation, we propose Test-Time Policy Optimization (TTPO), an asymmetric objective that distills agreeing rollouts via OPSD and penalizes disagreeing rollouts with Grouped RL. Token-level selection further refines both branches: distillation down-weights already-converged positions, while RL penalizes only confident errors. Both updates remain well-grounded even under frequent pseudo-label errors, and majority-vote routing yields tighter self-supervision as the model improves. Without any labels, TTPO matches label-supervised OPSD on five competition-level benchmarks, raises Qwen3-1.7B from 38.0% to 45.2% in TTT, yields +25.2% to +36.4% without thinking, and shows strong cross-task generalization.

State-of-the-art action-conditioned video models are typically restricted to a single robot embodiment, preventing them from leveraging the vast corpus of heterogeneous video data that contains rich signals for learning generalizable physics. To bridge this gap, we introduce CLAP, a framework for cross-embodiment action-conditioned video generation capable of being trained on diverse, internet-scale videos across human and robotic agents. CLAP is grounded in the insight that universal physical laws govern spatiotemporal dynamics regardless of the actor. However, cross-embodiment learning is non-trivial because action representations vary sharply across robot platforms and are typically absent in human videos. CLAP addresses this fundamental challenge through the following core contributions. First, CLAP reconciles disparate action spaces using end-effector poses, language instructions, and latent actions. Second, to resolve their individual limitations, CLAP introduces a curriculum-based cross-embodiment learning recipe that first learns foundational physical priors across unlabeled video data using latent actions and subsequently grounds them in end-effector action spaces for zero-shot deployment to real-world tasks. Crucially, CLAP approaches or surpasses state-of-the-art single-embodiment video models in challenging environments like DROID. These performance advantages compound via few-shot adaptation to establish a novel paradigm for training single-embodiment video world models. Ultimately, CLAP delivers the most comprehensive suite of action-conditioned video world models to date - spanning diverse action-conditioning spaces (end-effector, language, and latent) and robot morphologies (including cross-embodiment, DROID, Bridge, bimanual YAM robots, and G1 humanoids). We open-source all code and models. Project Website at https://omni-clap.github.io .

To improve large language models' ability to resolve real-world software issues, prior work has focused on constructing large-scale agent trajectory datasets and performing supervised fine-tuning (SFT) on successful trajectories. However, task success does not guarantee high-quality supervision: successful trajectories may still contain ineffective, redundant, or risky steps. Directly using such trajectories for SFT can introduce noisy supervision and encourage models to imitate undesirable problem-solving behaviors. Therefore, we propose SWE-Prime, a multi-granularity, two-stage SFT data selection method that progressively filters training data at the trajectory and segment levels. Specifically, the first stage performs trajectory-level screening based on process quality, result quality, and data representativeness, selecting a high-quality and representative subset of successful trajectories. The second stage performs segment-level selection by grouping consecutive steps into semantic segments and assessing each segment based on its contribution to the final solution, learnability, and potential risks. During SFT, all segments remain in the sequence to preserve context, while only selected segments contribute to the loss computation. Experiments on SWE-Bench Pro and SWE-Bench Verified show that training on the 10% trajectory subset selected by SWE-Prime outperforms training on the full resolved dataset, yielding relative performance gains of up to 12.2% and 24.2%, respectively.

Today, we're releasing Gemini Omni 1.1 Flash (gemini-omni-1.1-flash)!

⏩ Scene extension to 40s)
🖼️ First & last frame transitions
📹 3-second video references
🎨 Cleaner motions
🖥️ 1080p & 4K support
💨 360p draft mode, lower cost

Live today in the Gemini API

Today, we're kicking off the first phase of the research preview for Model Hardware Standard (MHS): a new standard for AI agents to safely operate physical equipment in scientific research and advanced manufacturing.

Read more: https://t.co/XQ2y9EW7Af

Runable says 60% to 70% of its 1 trillion-plus token usage in the last 90 days came from paying customers.

Startup investors put more than $546 million in disclosed dollar-denominated capital, plus €11.6 million, into the 10 most significant qualifying rounds identified in the latest 16-hour window. The biggest checks went to autonomous freight, AI data-center power management, and consumer AI, but the broader signal is more interesting: investors are increasingly funding technology that connects

Robot bodies are waiting for their AI brains to catch up.

Hot Chips 2026 sees Google discussing its new eighth-generation TPU family for the technical crowd. One of the only hyperscalers to develop its own training hardware, the company has developed the TPU 8t for training, as well as the TPU 8i for inference