Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…
gentic.news

AI Research

Breaking AI research news: latest papers from arXiv, NeurIPS, ICML, and top labs. Track transformer architecture advances, reasoning breakthroughs, and scientific discoveries in machine learning and artificial intelligence.

AI ResearchFunding & BusinessProducts & LaunchesBig TechStartupsOpen SourcePolicy & EthicsOpinion & Analysis
SMAC-Talk: StarCraft Benchmark Tests LLM Agents Against Deceptive Allies
AI Research
70

SMAC-Talk: StarCraft Benchmark Tests LLM Agents Against Deceptive Allies

SMAC-Talk extends StarCraft Multi-Agent Challenge with natural language communication, testing LLM agents against deceptive allies. Qwen3.5 models benchmarked; no model exceeds 72% win rate.

arxiv.org/Jun 5, 2026/3 min read/Multi-Source
ai safetyai agentsbenchmarks
Two AI model performance comparison charts side by side, one labeled Nemotron 3 Ultra and the other GPT-5.5, with…
AI Research
85

Nemotron 3 Ultra matches GPT-5.5 on physics test at 10X lower cost

Nemotron 3 Ultra matched GPT-5.5 on a physics test at 10X lower cost ($0.051 vs $0.57), highlighting MoE efficiency.

x.com/Jun 5, 2026/3 min read
moeai modelsinference
Anthropic engineer reviewing code on dual monitors with Claude AI interface displaying merged code contributions and…
AI Research
99

Anthropic: Claude Authors 80%+ of Code, Task Length Doubling Every 4 Months

Anthropic reports Claude authors 80%+ of code; task-length capability doubles every 4 months. Mythos Preview works 16+ hours autonomously.

x.com/Jun 4, 2026/3 min read
ai capabilitiescode generationclaude
A computer screen displays a complex web page with hidden code snippets highlighted in red, while a robotic hand…
AI Research
100

DeepMind paper: hidden web content hijacks agents 86% of the time

DeepMind catalogues 6 attack types where hidden web content hijacks AI agents up to 86% of the time, reframing safety from model alignment to environment trust.

x.com/Jun 4, 2026/3 min read/Multi-Source
research paperai securityautonomous agents
Two AI agent icons face a regulatory checklist document, with a 48.3% coverage score displayed above, representing…
AI Research
88

Ontology-Grounded AI Agent Testing Hits 48.3% Regulatory Coverage vs.

Ontology-grounded AI agent testing achieves 48.3% regulatory coverage vs. 33.1% baseline in 1800-scenario pilot. Coverage advantage over RAG not robust after Bonferroni correction.

arxiv.org/Jun 4, 2026/3 min read/Widely Reported
ai agentsregulatory complianceenterprise ai

Law Profs Prefer AI Answers 75% of Time in Stanford Study

AI Research
85

Law Profs Prefer AI Answers 75% of Time in Stanford Study

Stanford researchers found law professors preferred AI answers 75% of time in blind legal analysis test, per @rohanpaul_ai.

x.com/Jun 3, 2026/3 min read
legalresearchai
A person wearing headphones works on a laptop displaying a waveform interface, with floating musical notes and…
AI Research
85

Google Releases Magenta RealTime 2 for Open-Weight Music Generation

Google released Magenta RealTime 2 on Hugging Face, the only open-weights model for real-time continuous music generation on device with ~200ms latency.

x.com/Jun 3, 2026/3 min read
hugging faceopen-sourceedge ai
A line graph showing METR 80% task horizons over time, with a blue curve rising from 2024 to 2026; a red dot marks…
AI Research
85

Superforecasters Predicted 3-4h AI Task Horizons by Year-End; Claude Hit It in May

Superforecasters predicted 3-4h METR 80% task horizons by year-end 2026. Claude Mythos hit that in late May, compressing the timeline by seven months.

x.com/Jun 3, 2026/3 min read
claudeai benchmarksanthropic
A sleek black text-to-speech device with glowing blue accents sits on a desk, a waveform display showing emotional…
AI Research
85

Miso One: 8B Open-Source TTS Hits 110ms Latency, Real Emotion

Miso One, an 8B open-source TTS model, achieves 110ms latency with emotional range. Weights are fully open-source for self-hosting, but no benchmark data is provided.

x.com/Jun 3, 2026/3 min read
open-sourcevoiceoverai audio
A line chart showing a sharp upward slope from below 10% to 70%, with a Google logo and text 'LEAP scaffold' above…
AI Research
85

Google LEAP Scaffold Lifts Lean-IMO-Bench One-Shot Solve Rate from <10% to 70%

Google's LEAP scaffold lifts Lean-IMO-Bench one-shot solve rate from <10% to 70%, solving all 12 Putnam 2025 problems.

x.com/Jun 3, 2026/3 min read
leanautomated reasoninggoogle

MiniMax M3: Sparse Attention, 1M Context, Multimodal via …

AI Research
95

MiniMax M3: Sparse Attention, 1M Context, Multimodal via Together

MiniMax M3 uses sparse attention for 1M context and multimodality, with Together AI serving fast inference.

x.com/Jun 3, 2026/3 min read/Multi-Source
context windowai modelssparse attention
Researchers analyze a flowchart showing structured EHR data from CLMBR-T-Base feeding into a frozen LLM via a…
AI Research
92

ChatHealthAI: EHR Foundation Model + Frozen LLM Hits 79.8% F1 on Length-of-Stay

ChatHealthAI aligns CLMBR-T-Base with a frozen LLM via a task-aware resampler, achieving 79.8% F1 on EHRSHOT length-of-stay prediction while enabling interpretable reasoning.

arxiv.org/Jun 3, 2026/3 min read/Widely Reported
llmaiclinical decision support
A sleek, futuristic server rack with blue glowing lights, representing Microsoft's MAI-Thinking-1 AI model, a 35B…
AI Research
100

Microsoft Unveils MAI-Thinking-1: 35B Active, 1T Parameters, 97% on AIME 2025

Microsoft's MAI-Thinking-1 hits 97% on AIME 2025 with 35B active params in a 1T MoE model, trained on 30T human tokens without distillation.

x.com/Jun 2, 2026/3 min read/Widely Reported
reasoningopen-source aimicrosoft
Line graph showing performance peaking at 4 agents then declining, with researchers analyzing results on a laptop
AI Research
100

Multi-Agent Systems Hit Diminishing Returns Past 4 Agents

Adding more agents to LLM-driven multi-agent systems degrades performance past a task-dependent optimum, with weaker models peaking at 4 agents and stronger ones at 2.

x.com/Jun 2, 2026/3 min read/Widely Reported
scaling lawsmulti-agent systemsllm research
Two robotic arms play chess, one holds a piece mid-move, while a glowing digital brain hovers above the board…
AI Research
92

New 474-Game Benchmark Reveals LLMs Collapse on Counterfactual Reasoning

New 474-game benchmark reveals LLMs fail on counterfactual reasoning, with larger drops than contextual perturbations. Highlights metacognitive gaps in agentic AI.

arxiv.org/Jun 2, 2026/3 min read/Widely Reported
ai safetyresearchbenchmarks
MiniMax social media post showing a 26% BU Bench improvement claim for embodied AI planning, with no paper or method…
AI Research
95

MiniMax Claims 26% BU Bench Gain, Details Scarce

MiniMax claimed 26% BU Bench improvement without paper or code. Unverifiable claim reduces credibility.

x.com/Jun 1, 2026/3 min read/Widely Reported
unverified claimsminimaxbenchmarks
NVIDIA CEO Jensen Huang holds a large GPU chip, with a glowing data center backdrop and a chart comparing Nemotron 3…
AI Research
87

NVIDIA Nemotron 3 Ultra: 550B Open-Weight Model Challenges GLM, Kimi

NVIDIA released Nemotron 3 Ultra, a 550B open-weight model claiming near-SOTA performance, competing with GLM-5.1 and Kimi K2.6. No benchmarks yet.

x.com/Jun 1, 2026/3 min read
open sourceopen-weight modelsnvidia
Chinese Huawei chairman Xu Zhijun at a tech event, gesturing as he thanks US sanctions, with a projected roadmap…
AI Research
85

Huawei Chairman Thanks US Sanctions, Claims 1.4nm Equivalent by 2031

Huawei chairman thanks US sanctions, unveils Tau Scaling Law targeting 1.4nm density by 2031 via signal-speed optimization, not transistor shrinking.

x.com/May 30, 2026/3 min read
us-china tech warsemiconductorshuawei
Line chart comparing open-weight and frontier AI model release dates, showing a four-month lag gap narrowing over time
AI Research
79

Open-Weight Models Trail Frontier AI by Four Months: EpochAI

EpochAI finds open-weight models trail frontier closed-source models by four months, a small gap reflecting rapid catch-up.

x.com/May 29, 2026/3 min read
open sourceepochaibenchmarks
A line graph comparing EFC and raw token counts against agent harness success rates, with EFC showing a tighter…
AI Research
88

Agent Harness Scaling: EFC Predicts Success at R2 0.99 vs 0.42

New research introduces Effective Feedback Compute (EFC), which predicts agent success at R2 0.99 vs 0.42 for raw tokens. Reallocating compute by EFC lifts success 3x at the same budget.

x.com/May 29, 2026/3 min read
scaling lawsresearchai agents
ByteDance Open-Sources BAGEL: 7B Multimodal Model for Image Gen, Editing, Understanding
AI Research
95

ByteDance Open-Sources BAGEL: 7B Multimodal Model for Image Gen, Editing, Understanding

ByteDance open-sourced BAGEL, a 7B multimodal model for image gen, editing, style transfer, and understanding under Apache 2.0.

x.com/May 28, 2026/3 min read
image-generationopen-sourcemultimodal-models
Researchers at University of Texas present AgingBench benchmark on a screen, with AI agent memory bars declining…
AI Research
100

AgingBench: AI Agents Lose Reliability Over Time & Memory Fails

UT Austin paper finds AI agents degrade over time via memory errors. Proposes AgingBench to measure reliability decay across sessions.

x.com/May 28, 2026/3 min read/Widely Reported
deploymentmemoryai agents
A person sleeping peacefully in bed surrounded by glowing digital circuits, representing AI memory consolidation…
AI Research
84

Sleep Phase Cuts Transformer Costs by Consolidating Memory

Paper proposes sleep phase to consolidate context into fixed-size memory, reducing inference cost while improving long-horizon task performance on GSM-Infinite.

x.com/May 28, 2026/3 min read
memory & contextlanguage modelsai research
3D wireframe model of a chair overlaid with depth estimation heatmap colors, with small icons for surface normal and…
AI Research
91

SpatialBench: New Benchmark Tests Foundation Models on 3D Tasks

SpatialBench, a new benchmark from ropedia_ai, evaluates spatial foundation models across 7 tasks and 5 datasets, testing depth estimation, surface normal prediction, and 3D object detection.

x.com/May 27, 2026/3 min read/Multi-Source
foundation-modelsbenchmarkscomputer-vision
A person holding a smartphone with a bounding box around a cat on the screen, illustrating real-time object detection
AI Research
85

RF-DETR Hits Hugging Face Transformers: SOTA Real-Time Detection

Roboflow's RF-DETR, a SOTA real-time detection model, integrated into Hugging Face Transformers, bridging DETR accuracy with real-time speed.

x.com/May 27, 2026/3 min read
open-sourcereal-time-inferenceroboflow
NVIDIA Vera CPU on a test bench with cooling hardware, surrounded by cables and monitoring equipment in a lab setting
AI Research
100

NVIDIA Vera CPU Benchmarks: 1.55x Faster Than Intel Xeon in Phoronix Tests

NVIDIA Vera CPU benchmarks show 1.55x performance over Intel Xeon 6980P and 10% over AMD EPYC 9575F, with 1.2 TB/s memory bandwidth.

x.com/May 27, 2026/3 min read/Widely Reported
hardwarebenchmarksnvidia
A hyperrealistic 3D-rendered human face with intricate skin pores and micro-details, illuminated by studio lighting…
AI Research
85

DeemosTech Rodin Gen-2.5: 10M-Polygon 3D GenAI in 4 Seconds

DeemosTech claims Rodin Gen-2.5 generates 10M polygon 3D models in 4 seconds with skin microstructures, but provides no benchmarks or technical details.

x.com/May 27, 2026/3 min read
deemostechgenerative models3d generation
Anthropic engineers discuss code on monitors in a modern office, with a diagram showing AI agent permission levels…
AI Research
94

Anthropic Sandboxing Agents by Capability Level

Anthropic sandboxes agents by capability level, limiting destructive actions as agents gain autonomy in Claude.

x.com/May 26, 2026/3 min read/Multi-Source
access controlagent frameworksanthropic

MiniMax M3 Sparse Attention: 15.6x Decoding Speedup at 1M…

AI Research
100

MiniMax M3 Sparse Attention: 15.6x Decoding Speedup at 1M Tokens

MiniMax M3 sparse attention achieves 9.7x prefilling and 15.6x decoding speedup at 1M tokens, reversing M2's full-attention stance.

x.com/May 26, 2026/3 min read/Widely Reported
open sourcemachine learningai
A diagram showing a knowledge graph with connected nodes representing entities and relationships, with a central…
AI Research
95

Zep AI's Graphiti: Agent Memory Without Schema Is Just Storage

Zep AI's Graphiti enforces Pydantic schemas on LLM entity extraction, preventing generic label collapse and enabling precise querying of agent memory.

x.com/May 26, 2026/3 min read/Widely Reported
knowledge graphsai infrastructureopen-source