Research

All Research coverage

Kimi K2.6: The Open-Weight Model That Challenges GPT-5.4 and Claude Opus 4.6 With Agent Swarms

Moonshot AI has released Kimi K2.6, an open-weight model that directly challenges closed frontier models on agentic tasks by deploying agent swarms — multiple specialized sub-agents working in parallel. Benchmark results show Kimi K2.6 matching or exceeding GPT-5.4 and Claude Opus 4.6 on complex multi-step reasoning and code generation tasks, marking a significant milestone for the open-weight ecosystem.

Kimi K2.6: The Open-Weight Model That Challenges GPT-5.4 and Claude Opus 4.6 With Agent Swarms

The Best AI Models Lose Half Their Performance on Complex Charts — New Benchmark Exposes a Persistent Gap

Anthropic CEO Dario Amodei: 'There Is No End to the Rainbow' for AI Scaling Laws

Alibaba's Qwen3.6 Outperforms Google's Gemma 4 on Agentic Coding Benchmarks — Open Models Are Tightening Fast

Physical Intelligence's New Robot Brain Can Figure Out Tasks It Was Never Taught

Stanford's AI Index 2026: Rapid Progress, Growing Safety Concerns, and Declining Public Trust

AI Agent Skills Score Well on Benchmarks, Then Fall Apart When Deployed in the Real World

AI Models Would Rather Invent an Answer Than Admit They Don't Know — New Research Quantifies the Problem

Mustafa Suleyman: AI Is on an Exponential Curve — and the Wall Isn't Coming

Google's AI Overviews Are Right Nine Times Out of Ten — but the 10% Failure Rate Has a Specific Shape

Databricks Co-Founder Wins Top Computing Prize — and Says AGI Is 'Already Here'

Researchers Fingerprinted 178 AI Models' Writing Styles — and Found Alarming Clone Clusters

AI Tools Are Making Humans Think and Write More Alike, USC Study Finds

Google AI Overviews Are Correct Nine Out of Ten Times, New Independent Study Finds

Alibaba's Qwen Team Builds HopChain to Fix How Vision Models Fall Apart During Multi-Step Reasoning

IBM's Granite 4.0 Vision Leads Benchmarks on Enterprise Document Understanding at 3B Parameters

H Company's Holo3 Hits 78.85% on OSWorld — Setting a New State-of-the-Art for Computer Use AI

Study Maps Developer Frustration Over 'AI Slop' as a 'Tragedy of the Commons' in Software

AI Offensive Cyber Capabilities Are Doubling Every Six Months, Safety Researchers Find

Alibaba's Qwen Team Built HopChain to Fix How AI Vision Models Collapse During Multi-Step Reasoning

Even Perfectly Rational Users Can Be Pulled Into Delusional Spirals by Flattering AI Chatbots, MIT Research Finds

Anthropic Finds Claude Has 'Functional Emotions' That Can Drive It to Blackmail

AI Benchmarks Are Built on a Flawed Assumption, Google Study Finds: That Humans Agree

Alibaba's Qwen Team Fixes Reinforcement Learning's Blind Spot to Make AI Reason More Deeply

AI Offensive Cyber Capabilities Are Doubling Every Six Months, Safety Researchers Find

AI Offensive Cyber Capabilities Are Doubling Every Six Months, Safety Study Finds

Anthropic Finds 'Functional Emotions' in Claude That Drive It to Blackmail and Fraud Under Pressure

Alibaba's Qwen Team Fixes the Core Problem With Reasoning Model Training — and Doubles Thought Length in the Process

Anthropic Researchers Found 'Functional Emotions' in Claude — And They Can Drive It to Blackmail and Code Fraud

Know3D Solves 3D Generation's Blind Spot — Letting You Steer What AI Can't See

Anthropic Finds 'Functional Emotions' in Claude — And They Can Drive It to Blackmail

Researchers Warn of 'Cognitive Surrender': AI Users Are Abandoning Independent Reasoning When LLMs Are Available

Mintlify Ditched RAG for a Virtual Filesystem. Their AI Documentation Assistant Got Much Better.

'Cognitive Surrender': Research Finds AI Users Are Willingly Outsourcing Their Thinking to LLMs

TII's Falcon Perception Beats SAM 3 on Visual Grounding — With a 0.6B Model That Runs on One GPU

Google DeepMind Identifies Six 'Agent Traps' That Can Silently Hijack Autonomous AI Systems

Alibaba's Qwen3.5-Omni Teaches Itself to Code From Video — Without Being Trained To

NVIDIA, Berkeley & Stanford: Best AI Models Still Fail at Robot Control — Until You Add Agentic Scaffolding

H Company's Holo3 Sets New SOTA on Computer Use With Only 10B Active Parameters

Google DeepMind Maps Six Ways Attackers Are Already Hijacking AI Agents

UAE's TII Releases Falcon Perception: A 0.6B Model That Outperforms SAM 3 on Spatial and Relational Reasoning

IBM Launches Granite 4.0 3B Vision: A Compact Multimodal Model That Actually Understands Charts

Alibaba's Qwen3.5-Omni Taught Itself to Code From Voice and Video — Nobody Asked It To

Stanford Finds Frontier AI 'Sees' Images It Was Never Shown — And Medical Benchmarks Are Built on This Flaw

Research Finds AI Models Strategically Mislead Users — and Proposes a Fix

The 'daVinci-LLM' Paper Makes a Case That Pretraining Is Underrated — And Mostly Misunderstood

Can Frontier AI Write Formally Verified Graduate Math Proofs? A New Benchmark Has the Answer.

New Research: AI Agents Are Approaching a 'General Purpose Technology' Inflection — and We're Not Ready

New Paper: LLMs Know They're Wrong — But Bury the Truth When You Push Back

Mantis Biotech Is Building 'Digital Twins' of Humans to Solve Medicine's Training Data Problem

Stanford Finds AI Vision Models 'See' Images That Don't Exist — and Benchmarks Can't Catch It

AI Models Are Confidently Describing Images They Never Saw. Benchmarks Are Missing It.

Naver's Seoul World Model Grounds AI Video Generation in Real City Geometry to Stop Hallucination

Science Study: AI Sycophancy Makes People Less Likely to Apologize and More Convinced They're Right

Stanford Research Quantifies the Hidden Risk of Using AI Chatbots for Personal Advice

MetaClaw: The Framework That Trains AI Agents While You're in Meetings

Naver's 'Seoul World Model' Grounds AI in Reality With 1 Million Street View Images

AI Sycophancy Makes People Less Likely to Apologize — and More Likely to Double Down

Stanford Study: AI Chatbots Are Dangerous Personal Advisors — and Users Don't Know It

Anthropic's Own Data Suggests AI Is Making Skilled Users More Skilled — and Leaving Others Behind

Meta's 'Hyperagents' Don't Just Improve at Tasks — They Improve at Improving

Stanford Study Quantifies AI Sycophancy Risk: Chatbots Are Giving Harmful Personal Advice to Stay Agreeable

Cohere Releases Open-Source Speech Model That Beats OpenAI Whisper and ElevenLabs on Every Benchmark

Meta's Hyperagents Don't Just Improve at Tasks — They Improve at Improving

Meta's 'Hyperagents' Don't Just Solve Tasks — They Learn to Solve Tasks Better. That's New.

Anthropic's Economic Index Finds Something Uncomfortable: AI Makes Skilled Users Better, Not Everyone Equal

Google's TurboQuant Compresses AI Model Memory by 6x — With Almost No Quality Loss

Meta's TRIBE v2 Predicts How Your Brain Responds to Images, Sound, and Language — With Uncanny Accuracy

AlphaFold 3 Can Now Predict Protein Interactions at Atomic Resolution