Latest AI News

blogs.nvidia.com/Jun 22, 2026/3 min read/Widely Reported

LANL Taps NVIDIA Vera CPUs for 7x Agentic AI Speed on Scientific Workloads

LANL selects NVIDIA Vera CPUs for three supercomputers, claiming 7x performance on agentic AI workloads over x86. Systems Mission, Vision, Veritas deploy by 2027.

hpcai agentsnvidia

Zhipu AI founder Tang Jie gestures during a conversation with Elon Musk, as a leaderboard shows GLM-5.2 ranked No. 2…

scmp.com/Jun 22, 2026/3 min read/Widely Reported

90

Zhipu GLM-5.2 Hits No. 2 Globally; Tang Tells Musk China Won't Wait Until

Zhipu's 744B-parameter GLM-5.2 ranks No. 2 globally on Code Arena. Tang Jie tells Musk China will match Fable 5 by end of 2026, not Q1 2027.

chinafundingbenchmarks

A robotic arm assembles a circuit board in a high-tech manufacturing facility, with sparks flying from a welding tool

Funding & Business

84

Robotics Startups Raise Record $18.8B in 2026, Led by Saronic and Neura

Robotics startups raised $18.8B in first half 2026, exceeding 2025's $15B total. Saronic and Neura Robotics led with $1.75B and $1.4B rounds.

news.crunchbase.com/Jun 22, 2026/3 min read/Multi-Source

venture capitalroboticsfunding

A dynamic dashboard with interconnected nodes representing multiple LLMs, coordinated by Sakana AI's Fugu…

the-decoder.com/Jun 22, 2026/3 min read/Widely Reported

85

Sakana AI's Fugu Orchestrator Matches Anthropic Fable 5 Without Using It

Sakana AI's Fugu orchestrator matches Anthropic's top models on benchmarks without using them, offering a hedge against vendor lock-in amid export controls.

startupsbenchmarksai models

A 3D spatial tree diagram with branching nodes and arrows illustrating hierarchical spatial reasoning, with…

AI ResearchBreakthrough

100

ByteDance Seed's SpatialTree Redefines MLLM Spatial Reasoning at CVPR 2026

ByteDance Seed's SpatialTree achieves 79.8% on SEAL-Bench, 12.4 points above GPT-4V, using hierarchical spatial decomposition. Open-sourced at CVPR 2026.

pandaily.com/Jun 22, 2026/3 min read/Widely Reported

bytedancecomputer visionai research

pandaily.com/Jun 22, 2026/3 min read/Widely Reported

95

Microsoft Ditches Unlimited Copilot Tokens, Taps DeepSeek V4 for Cost Cuts

Microsoft switched Copilot Cowork to usage-based pricing, adopting DeepSeek V4 to cut inference costs by ~40%. The move breaks Microsoft's exclusive reliance on OpenAI for first-party AI.

open-sourcemicrosoftpricing

A graph with a rising red line crossing a dashed threshold labeled CB-2, with a yellow intermediate zone marked…

lesswrong.com/Jun 22, 2026/3 min read

71

Anthropic's CB-2 Gap Shows Biorisk Thresholds Need Intermediate Warning Levels

Anthropic deployed protections for Mythos 5 despite CB-2 not being crossed. The gap reveals a structural bias in biorisk thresholds that intermediate warning levels could fix.

anthropicai safetybiorisk

A complex flowchart of AI pipeline nodes and cost arrows, with magnifying glass highlighting hidden token fees

pub.towardsai.net/Jun 21, 2026/3 min read/Multi-Source

83

Thinking Tokens Drive Hidden Inference Costs in Agentic Pipelines

Thinking tokens from OpenAI, Anthropic, and Google models are priced at output rates, silently inflating costs 5x–10x in agentic pipelines. Google's 80% price cut threat exposes a structural asymmetry between startups and tech giants.

agentic aiaiinference

A laptop screen displays the HuggingFace dataset page for LOCUS-v1, showing 2.2M US laws, with code and data tables…

x.com/Jun 21, 2026/3 min read

89

LOCUS-v1: 2.2M US Laws Hit HuggingFace via AI Pipeline

LOCUS-v1, a dataset of 2.2M US laws built via AI pipeline, released on HuggingFace. First comprehensive legal database of its kind, but quality and validation metrics remain undisclosed.

hugging-facelegal-techdatasets

Cursor Trains GPT-Size Model with 10-20x Compute

x.com/Jun 21, 2026/3 min read

91

Cursor Trains GPT-Size Model with 10-20x Compute

Cursor trained a GPT-size model from scratch with 10-20x more compute, announced at Compile. The move shifts from fine-tuning to pretraining for code generation.

startup infrastructureai coding assistantsmodel training

AWS executives on stage at a New York summit presenting Continuum and Context services, with a large screen showing…

the-decoder.com/Jun 21, 2026/3 min read/Widely Reported

AWS Launches Continuum and Context to Fix Agent Blind Spots

AWS launched Continuum and Context to fix AI agent security and context gaps. Both services automate vulnerability handling and knowledge graph construction.

ai agentsawsenterprise ai

A Miami startup's LLM inference dashboard shows 12 million tokens processed for $8, compared to $2,600 on Claude…

AI ResearchBreakthrough

90

Miami Startup Claims 12M-Token LLM Inference at $8 vs. $2,600 on Claude

Miami startup claims 12M-token LLM inference for $8 vs. $2,600 on Claude Opus 4.6. No paper or benchmarks released yet.

pub.towardsai.net/Jun 21, 2026/3 min read

ai startupsllm inferenceanthropic

A diagram shows multiple robot agents connected by arrows, with a central meta-skill node labeled 'orchestration'…

x.com/Jun 20, 2026/3 min read

80

Meta-skill evolution lets multi-agent systems self-improve without retraining

Multi-agent systems can improve orchestration by evolving a meta-skill via RL on interactions, without retraining agents. Demonstrated on a simulated benchmark.

multi-agentmeta-learningreinforcement learning

Rows of humanoid robots working in a warehouse, assembling products on a conveyor line, with a few human workers…

x.com/Jun 20, 2026/3 min read

75

Figure robot count surpasses human headcount for first time

Figure's robot count surpassed its human headcount for the first time, signaling a shift from R&D to deployment. Exact numbers were not disclosed.

deploymentroboticsmilestone

A macOS window showing the OpenAI Codex app interface with a mouse cursor hovering over a Record button, suggesting…

the-decoder.com/Jun 20, 2026/3 min read/Multi-Source

94

OpenAI Codex Record & Replay: One-Shot Workflow Recording Becomes Reusable Skill

OpenAI's Record & Replay lets Codex learn a workflow from one demo and repeat it autonomously. The feature is blocked in the EU, UK, and Switzerland.

ai agentscodexworkflow automation

A bar chart comparing Zhipu GLM 5.2 and Claude Fable 5 scores on web design benchmarks, with GLM 5.2 leading in…

pandaily.com/Jun 20, 2026/3 min read/Widely Reported

Zhipu's GLM 5.2 claims Design Arena's top HTML spot with Elo 1,360 — edging a hobbled Claude Fable 5

Zhipu AI's 753-billion-parameter open-weight model GLM 5.2 topped the Design Arena HTML benchmark with an Elo score of 1,360, edging Anthropic's Claude Fable 5 (1,350). The win coincides with a Commerce Department export-control order that pulled Fable 5 from non-US users, and GLM 5.2's API pricing

anthropicchinese aibenchmarks

Big Tech

the-decoder.com/Jun 20, 2026/3 min read/Widely Reported

OpenAI Q1 revenue triples to $5.7B but burns $3.7B

OpenAI Q1 revenue tripled to $5.7B but cash burn also tripled to $3.7B. Stock compensation hit $2.3B. With $73B reserves, no immediate capital need, but a price war with Anthropic looms.

ai startupsfinancialsopenai

A line graph showing model performance scores from multiple AI labs, with ten colored trend lines clustered closely…

x.com/Jun 19, 2026/3 min read

75

The AI benchmark gap has collapsed: top 10 labs now separated by just 44 Elo points

Chatbot Arena Elo scores and Artificial Analysis data confirm that the top 10 AI labs are now clustered within 44 Elo points — the narrowest spread on record. Stanford HAI's 2026 AI Index corroborates the trend: leading frontier models are separated by as little as 3 percentage points on most benchm

industry trendsmodel performanceai competition

Norwegian flag on a school desk next to an open textbook and pencil, symbolizing the education policy change banning…

the-decoder.com/Jun 19, 2026/3 min read/Widely Reported

95

Norway Bans AI Tools for Under-13s, Pointing to Record-Low PISA Scores Since 2015

Norway will prohibit generative AI tools in grades 1-7 from late August 2026, citing falling PISA scores since 2015. Secondary students may use AI only under supervision. The policy extends an earlier smartphone ban that demonstrably improved grades and reduced bullying, and is backed by planned leg

ai policyeducationregulation

John Jumper in a suit speaking at an event, with a presentation screen behind him, likely discussing AI research…

Big Tech

100

Google DeepMind loses its third senior AI researcher in months as Nobel laureate John Jumper joins Anthropic

Nobel laureate John Jumper, DeepMind Director and VP Engineering Fellow who co-created AlphaFold, has left Google after nine years for Anthropic. The move follows Noam Shazeer's exit to OpenAI two days earlier — less than two years after Google paid $2.7B to reacquire him — and David Silver's Januar

the-decoder.com/Jun 19, 2026/3 min read/Widely Reported

anthropictalentdeepmind

UnitedHealth executives in a boardroom reviewing a large AI dashboard showing automated call metrics and medical…

Startups

bloomberg.com/Jun 19, 2026/3 min read/Widely Reported

UnitedHealth Bets $3B on AI Agents to Fix the Denial Machine It Built

UnitedHealth Group committed $3 billion to AI agents that call doctors, read charts to nurses, and process claims — a bet that the insurer that drew fury over algorithmic denials can use the same class of technology to restore trust. Under new CEO Stephen Hemsley, the company targets a 30% cut in pr

voice aiai agentshealthcare

A person using a laptop with ChatGPT interface open, surrounded by colorful AI-related graphics and charts…

AI ResearchBreakthrough

95

OpenAI shows small doses of beneficial-trait RL improve 44 of 53 safety benchmarks — and the gains generalize

OpenAI researchers Jagadeesh, Saab, Singhal et al. published findings on June 18 showing RL training on traits like honesty and corrigibility improved 44 of 53 safety benchmarks. Gains generalized across domains not used in training, and the model resisted harmful fine-tuning better than the baselin

the-decoder.com/Jun 19, 2026/3 min read/Widely Reported

alignmentai safetyreinforcement learning

A large language model interface displays Qwen 2.5 7B with a near-constant confidence score of 0.856, while…

arxiv.org/Jun 19, 2026/3 min read/Widely Reported

Qwen 2.5 7B Expresses Near-Constant Confidence Whether It Is Right or Wrong, Study Finds

A June 2026 arXiv preprint from University of Minnesota researchers tested Qwen 2.5 7B on structured clinical prediction data and found its verbalized confidence scores are essentially uninformative -- clustering between 0.856 and 0.937 no matter how well or badly the model performs. Combining SHAP-

researchsafetytabular data

AI Generates Chest X-Rays Clinicians Cannot Tell Apart From Real Ones

arxiv.org/Jun 19, 2026/3 min read/Widely Reported

85

AI Generates Chest X-Rays Clinicians Cannot Tell Apart From Real Ones

RadiT XL, a 1.3B-parameter rectified flow transformer trained on 1.2 million chest radiographs, produces synthetic images that clinical experts cannot reliably distinguish from real ones — a milestone that could break the data bottleneck limiting medical AI fairness and generalization.

medical imagingai modelsgenerative ai

A sleek ChatGPT interface on a digital screen displays a medical query with a detailed response, suggesting a health…

the-decoder.com/Jun 18, 2026/3 min read/Widely Reported

OpenAI Says GPT-5.5 Instant Beats Doctors on Health Accuracy — But It Designed the Test

OpenAI's GPT-5.5 Instant model reportedly outperformed doctor-written health responses across accuracy, clarity, and completeness in the company's own HealthBench evaluations, cutting flagged factuality errors by 71% over two months. The catch: OpenAI built the benchmark, organized the physician pan

chatgptopenaigpt-5.5 instant

White House Forced Anthropic to Cut SK Telecom Access, Triggering Model Shutdown

the-decoder.com/Jun 18, 2026/3 min read/Widely Reported

White House Forced Anthropic to Cut SK Telecom Access, Triggering Model Shutdown

White House forced Anthropic to cut SK Telecom access over China ties, then shut down Mythos and Fable 5 after security flaws emerged.

anthropicnational securityexport controls

OpenAI Can Predict Model Failures via Past Chat Replay

x.com/Jun 18, 2026/3 min read/Multi-Source

100

OpenAI Can Predict Model Failures via Past Chat Replay

OpenAI can estimate model failures by replaying past chats, enabling proactive error detection without new labeled data. No benchmark numbers disclosed.

ai safetyresearch

A sleek spa interior with warm lighting, a person relaxing in a circular hot tub, and a high-tech ultrasound scanner…

x.com/Jun 18, 2026/3 min read/Multi-Source

83

Midjourney Plans 60-Second Ultrasound Spa in SF by 2027

Midjourney plans a 2027 SF spa with 60-second ultrasound scans, aiming for 100x faster than MRI.

healthmidjourneyspa

A row of Iluvatar CoreX AI processors installed in a data center server rack, with cooling fans and blue LED lights…