Translate
Live intelligence

Latest AI updates

1,714 updates
AllModelsProductsIndustryResearchTutorialOpinion
Models

Ripple-65B Preview: Apache 2.0 Open Model Tops US Open Models on TerminalBench 2.1

Ricursive Intelligence has released Ripple-65B Preview, a 65-billion-parameter open-weight model under Apache 2.0. The company claims it achieves the top score among open-source U.S. models on the TerminalBench 2.1 coding benchmark, with a total R&D cost of $32 million. However, on other benchmarks, Ripple trails the newly released western open model Inkling 2 Small.

the-decoder27m agoScore 70
open-source AI modelscoding benchmarksmodel performance comparison
Products

SageMaker Inference now supports Multi-Instance GPU to maximize GPU utilization for model endpoints

AWS SageMaker Inference now supports NVIDIA Multi-Instance GPU (MIG) partitioning, allowing a single physical GPU to be divided into isolated partitions for deploying multiple models or right-sizing capacity. This feature works with inference components, enabling hardware-enforced isolation, predictable latency, and no cold starts. Benchmarks show a partition of an H100 outperforms a full A100 in throughput and cost for many workloads.

AWS Machine Learning Blog2h agoScore 79
Amazon SageMakerGPU utilizationModel inference
Products

Zyphra ZYFA: FP8 FlashAttention for AMD MI355X, Up to 1.76x Faster

Zyphra released ZYFA, an open-source FP8 and BF16 prefill attention kernel library for AMD Instinct MI355X GPUs. It achieves 1.16-1.76x speedups over AMD's AITER on real model activations, with lower FP8 error. The performance gains come from a lean softmax approximation and a two-wave ping-pong schedule that better overlap matrix and vector operations.

MarkTechPost2h agoScore 77
AMD Instinctattention kernelsFP8 inference
Products

Serverless MCP server for AI agent data lake access on Amazon ECS Fargate

AWS engineers built a serverless MCP server on Amazon ECS Fargate to give AI agents centralized access to enterprise data lakes. The architecture uses a metadata-driven catalog, so adding a new dataset requires only a JSON file, not custom code. It supports multi-tenant access with centralized security and scales automatically for bursty AI workloads.

AWS Machine Learning Blog3h agoScore 74
Model Context ProtocolAmazon ECS FargateData lake access
Tutorial

Govern AI agent access with Okta Cross App Access and Amazon Bedrock AgentCore – Part 1

This post demonstrates how to govern AI agent access using Okta Cross App Access (XAA) and Amazon Bedrock AgentCore. It details a three-token chain that enforces per-user authorization, with Okta as the governance plane and AgentCore Identity as the execution plane. The solution includes registering agents in Okta, configuring AgentCore Identity, and enforcing user authority in MCP servers. Performance tests show Okta revocation takes effect within seconds and AgentCore Identity adds minimal latency.

AWS Machine Learning Blog3h agoScore 74
AI agent securityidentity and access managementAmazon Bedrock
Industry Editor selected

Greenlight Financial builds autonomous SOC on Amazon Bedrock AgentCore

Greenlight Financial developed Luna, a tiered multi-agent system on Amazon Bedrock AgentCore, to triage security and risk cases. According to the case study, Luna autonomously resolves more than 90% of up to 4,000 monthly cases and escalates complex cases to human analysts. The system uses AgentCore runtime and memory, more than 30 specialized agents, episodic and long-term memory, and a case-learnings pipeline, and has expanded from SOC triage into financial risk assessment.

Why it matters: It provides a concrete production architecture and reported operational metrics for deploying multi-agent AI in a regulated security operations environment.

AWS Machine Learning Blog3h agoScore 82
Amazon Bedrock AgentCoreGreenlight Financialautonomous SOC
Products Editor selected

Taxbit Assist: A grounded, role-aware AI agent on Amazon Bedrock AgentCore

Taxbit built a permission-aware, citation-grounded AI assistant inside its digital-asset tax compliance dashboard using Amazon Bedrock AgentCore. The post details its architecture: Strands Agents SDK on AgentCore runtime, Bedrock Guardrails screening, a curated Bedrock Knowledge Bases index backed by Amazon S3 Vectors, Cohere embeddings and reranking, and per-session isolation. It also shares retrieval lessons such as excluding marketing pages, prepending document scope to chunks, and enforcing citation eligibility in code, plus continuous production evaluation and human-corrected answers.

Why it matters: It offers a concrete, production-tested case study of retrieval, grounding, identity binding and evaluation practices for shipping a regulated-domain AI agent.

AWS Machine Learning Blog3h agoScore 81
AWSAmazon BedrockTaxbit
Tutorial Editor selected

AWS sample runs Claude Code as a remote coding agent on Bedrock AgentCore

An AWS Machine Learning Blog post describes a sample that runs Claude Code as a remote agent on Amazon Bedrock AgentCore, with each task executed in an isolated microVM session inside the customer's AWS account. The workflow queues prompts, applies Amazon Bedrock Guardrails screening, clones a registered GitHub repository, plans and codes changes with Claude models, runs repository tests, and opens a pull request for human review. It adds security controls including short-lived GitHub tokens through AgentCore Identity, private VPC networking, least-privilege IAM, and a rule that the agent never merges changes. The sample supports submission from a web console or a Claude Code plugin and limits review cycles to three attempts.

Why it matters: It gives developers a concrete, security-focused implementation pattern for delegating coding tasks to remote AI agents while retaining human review and AWS-native controls.

AWS Machine Learning Blog3h agoScore 80
Amazon Bedrock AgentCoreClaude CodeAI coding agents
Industry

NVIDIA Open Source Tools Accelerate 6G Research with AI-Native Networks

NVIDIA highlights how its open source tools, including AI Aerial and Sionna, are being adopted by researchers to build AI-native 6G networks. The tools enable faster simulations, AI-powered signal reception, and energy-efficient network management, with adoption growing 10x and over 1 million downloads. Examples from King's College London, SoftBank, ETH Zurich, University of Leeds, and CTTC demonstrate concrete improvements in robot task efficiency, uplink throughput, power savings, and simulation speed.

NVIDIA Blog4h agoScore 79
6G networksAI in telecomNVIDIA Aerial
Research Editor selected

ToyBlocks: A controlled world for robot learning and evaluation

ToyBlocks is a new simulated environment designed for robot learning and evaluation, offering a controlled world with customizable tasks. It provides a platform for benchmarking and developing robotic manipulation algorithms. The blog post introduces the environment and its features, aiming to facilitate reproducible research in robotics.

Why it matters: This is a concrete, recent AI development that provides a new tool for robotics research, offering practical value to researchers and developers in the field.

Hugging Face Blog6h agoScore 82
robot learningsimulation environmentsbenchmarking
Research

METR audit suggests Anthropic may have crossed its own AI acceleration warning threshold in July

Nonprofit METR audited Anthropic's internal research data over eleven weeks and found that the pace of model development has roughly doubled, with about 80 percent of the speedup driven by AI-assisted research. Anthropic redacted the most informative evidence, so the public must rely heavily on METR's judgment. The findings suggest Anthropic may have crossed its own AI acceleration warning threshold back in July, though some internal voices remain skeptical.

The Decoder6h agoScore 68
AnthropicAI safetyAI development speed
Products

Thinking Machines' Inkling 2 Small claims top spot among western open-weight models

Thinking Machines, the startup led by former OpenAI CTO Mira Murati, is releasing Inkling 2 Small, a mixture-of-experts model with 594 billion parameters. The company claims it is the strongest western open-weight model, with improved agentic performance. However, this performance comes with significantly higher token usage.

The Decoder6h agoScore 70
open-weight modelsAI startupsmodel performance
Products Editor selected

Thinking Machines Unveils Inkling 2 Small: 17B-Active MoE Trained in 68 Hours

Thinking Machines released Inkling 2 Small, a 594B-total/17B-active multimodal MoE, trained from scratch in 68 hours on 8.4T tokens (NVFP4). It scores 31 on the Artificial Analysis Intelligence Index, beating many same-size Chinese models and ranking high among open-weight models. The model is available on Tinker with Apache 2.0 weights promised next week after safety testing.

Why it matters: This is a major open-weight model release with unprecedented training efficiency and competitive performance, offering concrete technical details and independent benchmarks.

latent9h agoScore 82
Thinking MachinesInkling 2 Smallefficient pretraining
Products

Humanoid robot cleaning service in San Francisco: a hands-on review

A journalist tests Tau Robotics' humanoid cleaning robot service in San Francisco, which is currently teleoperated by humans but collects data to train autonomous models. The robot can vacuum and mop autonomously but needs human help for tasks like folding laundry. The service costs $30 per hour for a single robot, and the robot is charming but slow and imperfect.

technologyreview10h agoScore 70
humanoid roboticsAI training datateleoperation
Industry Editor selected

Palo Alto Networks in talks to buy AI agent security startup Clutch for $500M

Palo Alto Networks is reportedly in discussions to acquire Clutch, an AI agent security startup backed by Lightspeed and SignalFire, for around $500 million. Clutch was last valued at $85 million, indicating a significant premium. The deal highlights growing demand for security solutions tailored to AI agents.

Why it matters: This is a concrete, recent M&A development in the AI security space, offering insight into market valuations and strategic moves.

techcrunch13h agoScore 83
AI agent securitycybersecuritystartup funding
Research

ReferClaw: Disambiguating User Intent for Egocentric Assistance

ReferClaw is an agentic framework for egocentric AI assistants that resolves ambiguous user queries by asking clarifying questions before answering. It uses three cooperating agents to identify candidate interpretations, request clarification, and generate grounded answers, guided by a cost-aware policy that balances information gain against user effort. Evaluations on augmented egocentric QA benchmarks show improved answer accuracy and zero-shot transfer across datasets.

arxiv15h agoScore 75
egocentric assistanceVideoQAuser intent resolution
Research Editor selected

MCPHallu: New Benchmark Exposes Hallucination-Prone Failures in MCP Agents

MCPHallu is a new benchmark designed to diagnose hallucination-prone failures in LLM agents using the Model Context Protocol (MCP). It includes 358 tasks across five domains, each with structural triggers for four failure subtypes: Branch Collapse, Unreachable Goal, Tool Misuse, and Context Forgetting. Evaluation of 14 frontier agents from eight vendors on 5,012 trajectories shows an average score of 0.642, with particular weakness on Unreachable Goal tasks (0.458 average, 23.2% pass rate). The benchmark enables subtype-level analysis, revealing that model rankings shift significantly across subtypes.

Why it matters: Provides a new diagnostic benchmark that goes beyond task completion to identify specific hallucination failure modes in MCP agents, offering actionable insights for improving agent reliability.

arxiv15h agoScore 85
Model Context ProtocolAI agent evaluationhallucination detection
Research

Equivariant Deep Learning in Direct-Product Space for Electronic-Structure Hamiltonian Prediction

Researchers propose DP-Learner, an equivariant learning framework that performs message passing directly on Hamiltonian-like operator blocks in direct-product space rather than direct-sum spaces of irreducible spherical tensors. The method introduces Lie-generator-based directional embedding and invariantly weighted left-right products and contractions intended to preserve equivariance to SO(3), SU(2), and spatial inversions without explicit Clebsch-Gordan decomposition. It is instantiated as a Transformer architecture and accompanied by MP-HAM-110K, a dataset of more than 110,000 material structures covering 68 elements with Hamiltonian and wavefunction data. Experiments on MP-HAM-110K and Materials-HAM-SOC report improved Hamiltonian prediction accuracy and inference efficiency over the state-of-the-art method.

arxiv15h agoScore 77
AI for scienceequivariant neural networkselectronic structure