Latest AI News

Get the fastest, most accurate AI news every morning

Total articles: 12,828|Added todayToday: 58

GitHub Releases Copilot Fusion

GitHub released Copilot Fusion in GitHub Copilot, coordinating models from Anthropic and OpenAI
Fusion cost $5.59 per task and scored 71 on Terminal-Bench 4.0
It used 7.2M tokens per task and scored 51 on FrontierCode and 57 on SWE-Atlas-QnA-AA

Artificial AnalysisRead more · 1 min →

Today's

Amazon Launches Alexa+ Early Access in Japan

Amazon began Alexa+ early access in Japan, offering it free to several million users for now.
After early access, Alexa+ is planned to cost ¥3,000 monthly; Prime members paying ¥5,900 annually can keep using it free.
Alexa+ supports natural conversation, personalized responses, smart home controls and planned links to Japanese services.

Impress WatchRead more · 3 min →

Today's

Gemini 4 Carbon Tops Intelligence Index v5

Gemini 4 Carbon scores 59 to lead Artificial Analysis Intelligence Index v5, taking top marks in 4 of 9 evaluations.
Carbon ranks second on the Coding Agent Index, 3 points behind Claude Opus 5.5, and second on two coding benchmarks.
Google is rolling Carbon out to selected users with a 50% launch discount and no confirmed end date.

Artificial AnalysisRead more · 2 min →

Today'sFinance

China AI Stocks Lose $480 Billion

Reuters reports a sharp slump in Chinese AI and semiconductor stocks and signs of an AI bubble starting to deflate
The STAR 50 fell more than 40% from its July peak, while market capitalization dropped by $480 billion
MiniMax and Z.ai face profitability concerns as price competition weighs on related stocks

AI Times KoreaRead more · 2 min →

Today'sInfra

Brazil Courts Data Centers With Renewable Power

After securing reelection, Lula said Brazil would use renewable energy to attract data centers and build its own AI.
Redata offers tax relief if facilities use renewable or low-emission power and reserve at least 10% of service capacity for Brazil’s domestic market.
Brazil’s energy plan projects data centers will use 30 TWh annually by 2036; renewables supplied more than 86% of electricity in 2025.

AI Times KoreaRead more · 2 min →

Today'sInfra

Nvidia and Oracle Expand OCI Partnership

Nvidia and Oracle launch Vera Rubin NVL72 and standalone Vera CPUs on OCI, with plans to expand OCI Superclusters to multiple gigawatts.
OCI will scale CPU and GPU resources independently as agentic AI workloads increase.
Nvidia claims Vera CPUs are up to 80% faster on agent-related CPU tasks than major x86 processors.

AI Times KoreaRead more · 2 min →

Today's

HCM Customers Set Conditions for AI Agents

Oracle AI World customers say process, data and legal approval come before HCM agents
IHG targets talent acquisition from 2027; Deutsche Bahn and UL Solutions are expanding Oracle HCM
Customers cite process redesign, legal review, testing, token costs and data preparation as adoption work

diginomicaRead more · 2 min →

Trending Keywords

Today'sInfra

Kolsky Outlines Enterprise AI Budget Priorities

Esteban Kolsky said enterprises’ 2027 AI budgets face tighter controls and scrutiny over returns and value.
At Connected Enterprise 2026, Kolsky identified six investment priorities: value, data, infrastructure, governance, economics and talent.
Kolsky said AI value may emerge around year two or three, with investments planned across five to 10 years.

Constellation ResearchRead more · 1 min →

Today'sHealthcare

Zocdoc Opens Appointment Booking to AI Agents

Zocdoc opens provider search and appointment booking to registered AI agents with patient consent
Google Gemini, Amazon Health AI and Instinct are already connected to Zocdoc
AI-agent traffic has grown nearly 30-fold since July, according to Zocdoc

Healthcare DiveRead more · 1 min →

Today's

Baseten Speeds Qwen3-ASR Batch Transcription

Baseten reports up to 7x higher Qwen3-ASR 1.7B batch transcription throughput than unmodified vLLM.
On a single RTX PRO 6000, hour-long audio reached 3,128x real-time speed at $0.0013 per audio hour.
At 64 concurrent requests, 5-minute clip latency fell from 50.4 seconds to 7.3 seconds.

BasetenRead more · 2 min →

Today's

LangChain Tests OpenWiki Agent Memory

OpenWiki made an agent roughly 5x more accurate on questions connecting information across documents than keyword search alone
An evidence graph raised accuracy 27%, doubled citation consistency, and cut median answer time 40%
Testing covered 111 insurance documents, 972 facts, 80 simple lookups, and 50 connected questions asked three times each

LangChainRead more · 2 min →

Today's

GitHub Reports 2026 Octoverse Growth

GitHub’s 2026 Octoverse report records sharp growth in activity associated with agentic engineering.
Merged pull requests tripled to 220 million monthly, while commits more than tripled to 7.38 billion.
New repositories doubled to 30.3 million monthly; report also breaks figures down by industry and repository topic.

Simon WillisonRead more · 1 min →

Today's

CodeCap-Bench Tests Code-Based Video Captions

Researchers introduce CodeCap-Bench to evaluate models’ ability to generate executable code captions for videos
The benchmark includes 323 videos in five structured 2D categories with fine-grained annotations and multi-level evaluation
Code captions outperform natural language on reasoning and reconstruction, while open-source models trail proprietary systems

HuggingFaceRead more · 1 min →

Today's

Study Finds Agent Memory Can Underperform RAG

University of Illinois researchers report many agent memory systems underperform a simple RAG baseline across benchmarks and model backbones
Paper frames memory’s net value as gains called memory benefit and losses called memory tax relative to raw-history retrieval
Commenters cite LongMemEval-s scores of 81.4 for two-round RAG, 76.6 for MemOS, and 78.2 for HippoRAG2

HuggingFaceRead more · 1 min →

Today'sHealthcare

Meta Supports AI Research at Five Universities

Meta will support research projects at five U.S. universities with models, researcher collaboration and selected funding.
Caltech projects span five sciences, including a benchmark built on more than 86,900 research-level math problems.
Ohio State and LSU teams plan clinical AI work, including heart-failure risk, radiology reports and infant monitoring.

Meta Superintelligence LabsRead more · 2 min →

Today's

Paper Proposes Multiscale Agent Training Method

Paper proposes Multiscale Policy Optimization for training long-horizon language-model agents
Experiments with 4B and 30B-A3B models report consistent gains over strong reinforcement-learning baselines
MPO combines Multiscale Advantage Estimation with turn-level importance sampling across search, tool-use and automation tasks

HuggingFaceRead more · 1 min →

Today'sCoding

Google Shows Agents CLI Evaluation Workflow

Google’s Agents CLI walkthrough demonstrates evaluation, regression tracking, model comparisons, scheduling and score-drop alerts.
A chef-prompt regression lowered the four-case mean from 4.0000 to 2.7500; three cases scored lower.
A model swap produced a 2.75 mean, while scheduled checks and ADK prompt optimization offer additional evaluation workflows.

Google DevelopersRead more · 2 min →

Today'sSafety & Ethics

Google Announces Gemini 4 Carbon

Google DeepMind announced Gemini 4 Carbon, reporting top cybersecurity benchmark results and long-horizon engineering capabilities.
Carbon scored 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench.
A limited cyber defender rollout begins this week; introductory prices are $2 per million input and $10 per million output tokens.

Google DeepMindRead more · 2 min →

Today's

Gemini Adds Immersive Explorations

Gemini app adds a library of interactive, visual explorations available through the Gemini student hub
Scenes cover Silk Road trade routes, jet engines and lightning, with layers from planets to atoms
Feature supports six languages; direct chat access and audio narration are planned

Google BlogRead more · 1 min →

Today's

AWS Adds Team Controls to QA Studio

AWS’s third QA Studio guide adds application health metrics and team-based access controls to Amazon Nova Act test automation.
Applications group use cases and suites; health scores range from 0–100, with trends across 7-day and 30-day windows.
Five cumulative roles govern access, while shared permissions are capped by each user’s own role.

AWS ML BlogRead more · 2 min →

Trending Keywords

Last 7 Days