Baseten BlogNVIDIA Nemotron 3.5 Super VL combines visual understanding, reasoning, and tool use, supporting reading documents, images, application interfaces, and videos, scoring 1190 Elo in the GDPval evaluation, higher than Nemotron 3 Super's 493. Developers can fine-tune it via Baseten Loops SDK and evaluate it using NVFP4 weights on Blackwell GPU.
NVIDIA BlogNVIDIA releases the open-source 120B parameter Nemotron 3.5 Super VL model, achieving the fastest task completion speed and highest accuracy in multimodal agent tasks; simultaneously launching Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB edge models.
Mistral AIMistral AI released the Timestral time series foundational model family, with the smallest model at only 209M, ranking first in three major benchmark tests (GIFT-Eval, TIME, and fev-bench), outperforming models with larger parameter sizes; this series includes four sizes ranging from 5M to 209M parameters, supports past and future covariates, runs efficiently on CPUs, and can be fine-tuned for specific domains via Mistral Forge.
Apple Machine Learning ResearchThe Selection-Based Structured Reasoning (SSR) framework restructures reasoning into selection rather than open-ended generation, selecting the most likely solution from predefined reasoning candidates before each action, significantly reducing reasoning costs. Validated using 2B and 4B models on seven multimodal search benchmarks, SSR achieved average success rates comparable to leading search agents of the same scale, reducing single-step reasoning latency by 90% and total reasoning latency per task by 28-54%.
Together AI BlogMeta Muse Spark Open has been launched on the Together AI platform, which is currently the most powerful US open-source heavy model, designed for long-running intelligent coding workflows that require the entire codebase to be stored in context
Modal Blog7575Meta's Muse Spark Open large model is now available on Modal Auto Endpoints, supporting intelligent coding and long-term tasks, with native multimodal understanding and a 1M token context window, capable of completing code writing, debugging, multi-step planning, sequential tool use, and fault recovery in extended workflows.
OpenHands BlogOpenHands introduces NVIDIA Nemotron 3.5 Super VL (120B parameter MoE architecture, 12B active parameters) into software engineering agents, supporting text, image, document, video input and text output
Prime Intellect BlogPrime Intellect and NVIDIA open-sourced 2500 enterprise-level training tasks for NVIDIA Nemotron 3.5 Super VL, fine-tuning through 100 GRPO steps in about 11 hours on 128 NVIDIA GB200 GPUs.
Google Developers BlogGoogle has launched the ADK-Recipes repository as a single source of truth for Agent platform content, including core patterns, plugin skills, and services, supporting end-to-end construction of long-term memory agents, one-click installation of retail vertical skills, and deployment to Cloud Run or Gemini Enterprise.
Liquid AI Blogd1-ultra is the newly released decision model by Liquid AI that supports multimodal inputs, outperforming d1 and Jev 1.14 in 112 evaluations, providing more reliable probability outputs. The model is based on a larger open-weight model and has been customized for improvements, excelling in knowledge-related tasks, priced at $0.042 per million input tokens, and can be accessed via the Liquid API.
Apple Machine Learning Research3434On ToolBench, Retrieve-then-Plan (RtP) outperforms Plan-then-Retrieve (PtR) in single-tool queries, while PtR performs better in multi-tool queries, this crossover holds true across all retrieval budgets (k=1 to 20), registry sizes (1k to 47k APIs), and model families (Gemini 2.5 Flash and gpt-oss-120b).
Apple Machine Learning ResearchOn a cluster of four Mac Studio machines, Muse Spark Open (816B parameters) generates at a speed of over 36 tokens/second. The Mac Studio M5 Ultra version offers 4.6x faster inference speed (TTFT) and 1.6x faster tokens/second compared to the M3 Ultra, benefiting from the GPU neural accelerator and 1.2 TB/s memory bandwidth.
Artificial Analysis ArticlesNVIDIA Vera Rubin NVL72 significantly outperforms GB300 NVL72 in the AA-AgentPerf benchmark for proxy workloads, reaching up to 5.7 times better per MW. Each Rubin GPU features 288 GB HBM4 memory, 19.2 TB/s memory bandwidth, 3.0 TB/s NVLink bandwidth, and 35 PFLOPS dense NVFP4 computing, supported by a CPU with 88 cores.
Simon Willison6565Claude Opus 5.5 Max developed the Tree Appreciator game in 3 hours and 7 minutes; it is a slow-paced spot-the-difference game that generates two 3D trees for players to find differences, featuring music, animations, and an achievement system. Players can drag and zoom to observe, use limited hints, and earn points, leaves, and specimen entries for each pair of trees completed.
Latent SpaceLeash Bio optimizes DNA encoding library (DELs) data collection by attaching dozens of barcodes on molecules, its AI agents are on par with Boltz on public benchmarks, and has released a verifiable scientific reasoning AI tool named Minos and an open-source dataset.
AI as Normal TechnologyBeyond Automation: A new research direction in AI argues that AI should not be limited to automating human labor but should explore new directions for enhancing humans and promoting human-machine collaboration. This perspective advocates for AI research to focus on new issues such as transparent agents, process tracking, and human-machine interaction design to push AI beyond pure technological automation and achieve broader applications.
Google ResearchGoogle has launched the Geospatial Reasoning intelligent agent, integrating Earth AI base models (including WeatherNext, PDFM, etc.) and data sources like Google Maps, handling planetary-scale geographical data through modular skills and planning capabilities.
Ant Group inclusionAI · Hugging Face new modelsinclusionAI/Ming-3.0-asr-tiny combines Ling-tiny-v3 language backbone and audio encoder, supporting single-round Chinese and English speech transcription and recognition of 15 dialects, improves professional term recognition rate through hot words or dialogue context, model parameter size not mentioned, evaluated in noisy speech and 5 application areas, provides PyTorch/Transformers inference examples and MIT license.
NVIDIA BlogMultiple cities in Europe, along with software, infrastructure, and integration partners, are using NVIDIA's accelerated computing to build physical AI agents and digital twin systems to optimize urban operational decisions. These applications can predict maintenance needs, guide urban planning, and enhance public service response capabilities, involving Belgium, Finland, France, Italy, Netherlands, Poland, Sweden, and the UK.
StepFun · Hugging Face new modelsLeapX Stars releases flagship model Step-5-Preview-NVFP4, with 600B total parameters and 27B active parameters, supporting up to 1M context windows, excels in software engineering, professional knowledge work, and financial fields, capable of processing text, images, and video, deployable via OpenAI-compatible API or offline generation.
Apple Machine Learning Research2727Whisper-small achieved an average F1 score of 0.798 across 14 language pairs in the CS-FLEURS benchmark test, surpassing CNN's 0.776 and HuBERT's linear baseline's 0.751. The method detects language switching directly from speech by identifying the language-selective hidden state dimension in pre-trained speech encoders, and only requires a small amount of code to convert the dataset to fine-tune decision rules.
Apple Machine Learning Research4040Apple Machine Learning Research has for the first time pre-trained a tri-modal masked diffusion model (MDM) from scratch on text, image-text, and audio-text data, analyzing multi-modal extension patterns.
n8n Blog7575n8n has integrated the newly released Gemini 4 Argon, allowing workflows to be built from other AI tools via the Model Context Protocol (MCP). The Assistant supports BYOK running on the user's own account Gemini 4 Argon, and the Agent can use Gemini 4 Argon for reasoning and decision-making.
Google DeepMindGemini 4 Argon has achieved major breakthroughs in physical understanding, code generation, and visual semantic reasoning, achieving a pure reasoning score of 87% on the Embodied Reasoning (ERQA) benchmark test, surpassing all leading models. Argon can be applied to all stages of the robot's lifecycle, including mobile manipulation practice using MuJoCo, zero-shot robot control, and closed-loop steering of video world models.
Google DeepMindGemini 4 Argon ranked first on the Vals Index, surpassing other cutting-edge models in coding, finance, law, cybersecurity, and science with its 175B parameter size; its reasoning depth and accuracy make it a powerful tool for solving complex engineering problems, optimizing business processes, and accelerating scientific breakthroughs.
Google AI BlogFeatured9090Google has released Gemini 4 Argon, opening it up to developers, businesses, and researchers, with API pricing at $2 per million input tokens and $10 per million output tokens, offering a 95% discount on cached input tokens.
Why it matters: Official pricing, baseline rankings, and multiple internal implementation cases are provided, allowing judgment on its position in coding and research workflows.
Liquid AI BlogLFM2.5-24B-A2B is a 24 billion parameter reasoning model that uses only 2 billion active parameters, suitable for fast intelligent agents on devices. The model competes with Qwen3.6-35B-A3B in five key benchmark tests, 1.7 times faster than similar models, and is open source.
Amp News7070Users can interact with Puck through MCP, brainstorm tasks in ChatGPT or Claude, and then send them to Puck for implementation and management. Puck knows the user's projects in Amp and routes them to the correct module, supporting direct viewing of Amp work progress in ChatGPT. MCP provides the puck tool for Amp interaction and the orbs tool to display the interaction view of running modules.
Prime Intellect BlogPrime Intellect acquires Moondream, its founder will join Prime Intellect to oversee reasoning and cross-modal model development. Moondream 3.1 performs excellently in object detection, aerial images, charts, and crowded scenes (such as shelves), processing 34 requests per second, 24 times faster than the next accurate open-source model, and costs only one-thousandth of Meta's new model Muse Wonder when achieving 80% accuracy.
Amazon ScienceAmazon Science proposes Priming technology, which reduces the pre-training cost of hybrid state space models by over 99% by leveraging pre-trained Transformer weights, enabling developers to explore problems that were previously too costly to investigate.
Import AI4242Humanity’s Sixth Sense review shows GPT 6 Astra scores 53.6 in intuitive visual reasoning tasks with a pass@1 rate, far below the human baseline of 93.1, with the open-source Chinese model GLM-5.3-flash scoring 26.8 and Kimi K3 scoring 25.5.
Baseten Labs7575Base Labs research found that iterative learning with the talkie-1930-13b-it model using sports facts from 2011-2019 leads to fact forgetting, with forgotten facts still present in model weights. Experiments show that early learned facts are difficult to recall after 20 writes and merges, and are covered by later facts, but the covered facts can be relearned, primarily occurring in the later MLP weights.
Anthropic ResearchClaude Code v2.1.159 is released, available for free on GitHub, supporting reasoning with 8k context length, and can run 9B models on 9 hardware types, but models below 30B parameters still error out during full replication.
Sakana AIMulti-Agent Self-Supervision (MASS) achieves recursive self-improvement through multi-agent self-supervision, enabling a 27B open-weight model to achieve scores 1.2-1.6 times that of the baseline model for each output token after the second round of synthetic open-ended research tasks.
Redwood Research7575Redwood Research publishes a paper empirically testing two distillation methods for AI safety: one for conviction (DFI) and one for capability (DFC). DFI distills untrusted models into weaker trusted models, aiming to transfer the misalignment but not the hidden capabilities of the teacher model, so that the student model can be audited to convict the teacher model; DFI is biologically tested on secret-keeping models in AuditBench, where the distilled student model acknowledges the frequency of quirks far higher than the original model biologically, indicating that quirks transfer faster than the tendency to hide quirks.
NVIDIA BlogNVIDIA Omniverse, combined with cutting-edge AI models like GPT-6 Astra, helps developers achieve simulation applications by guiding AI agents to complete asset assembly, physical connections, and rendering checks through natural language instructions.
AWS Machine Learning BlogAmazon SageMaker HyperPod provides centralized authentication through AWS IAM Identity Center, uses the SageMaker AI domain to provide a customized user experience, combines Kubernetes namespaces to achieve workload isolation, and leverages HyperPod task governance for fair resource allocation, supporting multiple teams in efficiently sharing a single HyperPod EKS cluster.
The Verge · AIGoogle has released a universal Gemini AI agent that can operate across applications and devices in work scenarios. The tool provides a single interface through the Gemini Enterprise app, supporting chat with the agent and task assignment, and can work directly within Workspace applications like Gmail, Drive, Docs while maintaining cloud context. It is currently only available to business customers for private preview, as Google had previously launched the consumer-focused Gemini Spark agent in May.
MIT Technology Review · AI6565Tesla's Optimus humanoid robot is expected to be sold to the public by the end of 2027, with each priced at about $20,000, but experts like Yann LeCun question whether its intelligence level will reach practical levels.
