2026

OpenAI: GPT‑5.6 Sol model finds critical flaw in PHP, which runs 70% of websites

Claude Mythos 5 designs quantum circuits that beat published records

Robotics

Wujie Dongli Open-Sources World Model MWA-Phoenix, Claims Top Spot on Six Benchmarks

Business·Jiqizhixin (机器之心)

DeepSeek open-sources its DeepInfer inference engine, and analysts calculate an 82% margin

DeepSeek has released its internal DeepInfer framework and the deepseek-recipe component under Apache 2.0. Third-party estimates put the gross margin of its AI…

Research·Jiqizhixin (机器之心)

RVCBench Tests 18 Voice Cloning Models on Noise, Compression, and Anti-Cloning Protection

UBC researchers present RVCBench at NeurIPS 2026: 18 voice cloning models tested on noisy recordings, phone channels, and "anti-clone" protection.

LLM·Anthropic Research

The Expectations Gap: 91% of Anthropic Employees and 39% of Americans Expect a Nobel-Level AI Breakthrough

Anthropic has published research on AI expectations: 91% of its employees and 39% of Americans expect Nobel-level discoveries by 2030. The gap in forecasts spans…

Business·Anthropic Research

Anthropic: AI Agents Now Outnumber Employees 6 to 1, and Junior Hiring Is Shrinking

Two Anthropic reports: by August 2026 the company has six times more AI agents than people, juniors make up a shrinking share of hires, and 91% of staff expect an…

01LILIE teaches image generation through language, cutting gFD from 66.84 to 53.01Oct 22, 2026·2 min 02Competence Over Accuracy: Why GPT and Qwen Lose Logic When Problems Get RephrasedOct 22, 2026·2 min 03A no-hair theorem for transformers: the prompt is forgotten as t^-α, α≈0.55–0.7Oct 22, 2026·2 min 04Huginn Beats Ouro on All 10 Benchmarks: The Design Space of Recurrent LLMsOct 22, 2026·2 min 05MOSS-Image: Open 5B Model Renders Long Text with 96.3% CoverageOct 22, 2026·2 min

Top themes

MIST shows video world models don't forget; they fail to carry out the command

"Wait" in Reasoning Models Turns Out to Be Inherited from Data, Not Discovered by RL

Business·Vercel Blog

Vercel Shop Reaches GA: Shopify Storefronts Learn to Sell to AI Agents

Research·Jiqizhixin (机器之心)

OpenAI's Interpretability Founder Says 90% of Neurons Have a Single Function

Hardware·Jiqizhixin (机器之心)

Kunlun Wanwei launches Matrix-Game 4.0: a video model that plays in 480p at 60 FPS on a single GPU

Kunlun Wanwei has released Matrix-Game 4.0, a game video model that outputs 480p at 60 FPS on a single consumer GPU. A game engine runs the world; a neural network…

LLM·DeepSeek Blog

DeepSeek releases V4.1-Flash-1017 and opens its Search API at $0.75 per 1000 calls

DeepSeek's updated Flash model, V4.1-Flash-1017, finishes tasks faster, writes shorter answers and is open source. Search and Fetch APIs cost $0.75 per 1000 calls.

Research·Jiqizhixin (机器之心)

Claude Traces the Origin of Five NIST Curve Seeds in Half an Hour

Anthropic says an internal Claude model worked out where the seeds of five NIST elliptic curves came from in under 30 minutes. The question had stayed open for more than 20 years.

HumanoidPose Estimates Humanoid Robot Pose From Video, Cutting Joint-Angle Error by 18–32%

Researchers Annotate 2,122 WildChat Conversations: 71.1% of Legal Chats Seek Personal Advice

StopLoss Cuts Off Doomed Coding-Agent Runs and Saves Up to 48% of Costs

Opinion·Жемал Хамидун · Авторская колонка

An LMS and CRM at 205,000 Lines of Code, Built by AI Agents: The Real Cost of Running It, Three Lead States, and One \r Character

Changing the definition of "unsafe" moves a guardrail score from 0 to 100 with 0.03% of weights

Anthropic to put 5% of its shares into a Public Benefit Fund, with the stake growing as the company grows

Research

Researchers Propose Anti-Persistence: LLM Agent Memory Without Stored Rules

Research

3D Scene Generation: DiT Works in a 3D Latent and Encodes a Scene in 1K Tokens

POPE audit: the choice of "yes/no" tokens shifts LLaVA-1.5 F1 by 6.13 points

Researchers tried to reproduce a VLM hallucination-mitigation method and found a flaw in the evaluation itself: wrong token IDs for "yes" and "no" shift POPE F1 by…

Parallel Refinement Instead of Diffusion: DRLMs Hit 93.9% on Sudoku-Extreme

Microsoft has introduced DRLMs — a parallel reasoning method that generates a full draft and iteratively refines it instead of running reverse diffusion. On three…

TokenWorth Teaches AI Agents to Forget: Best Results on Three Environments With the Fewest Tokens

Researchers introduce TokenWorth, a lightweight agent memory system that scores the "worth" of memories and outperforms more complex baselines on three…

Ropedia W0 trained on 10,000 hours of human video and 3,000 hours of robot data

Ropedia W0 is a world model for walking and manipulation in both humans and robots. Once distilled, it makes minute-long predictions in real time and runs in VR.

01Anthropic: junior share of hires fell from 27% to 9%, while agent hours outpaced human hours 6xOct 22, 2026·3 min 02TokenGate speeds up LLM prefill by up to 11.3x with learnable token pruningOct 22, 2026·2 min 03Vercel sets deadlines for vendors: static keys must go by 2027Oct 22, 2026·3 min 04T2Quant Reallocates Precision Across Diffusion LLM Steps: Up to +6.7 ppOct 22, 2026·2 min 05RIDAR catches integrity violations by AI scientists: 3964 cases across 300 runsOct 22, 2026·2 min

Top themes

FilmFlow: Four Agents Build 3D Camera Trajectories from Text and Video

LLM·Hugging Face Daily Papers

BrickWright Teaches Coding Agents to Build with LEGO: 100% Valid Builds vs. 9–69%

A Tiny 856K-Parameter Policy Speeds Up the Diffusion LLM LLaDA-8B

LLM·Hugging Face Daily Papers

IQuest-Q1: An Open 320-Billion-Parameter MoE Model for Agentic Coding in the CLI

Naive RAG Beat Agentic Memory Systems in Six of Seven Tests

Raw-RAG, which uses dense retrieval and reranking over raw conversation history, took first place in six of seven setups and beat the best agentic memory by…

AgentVQA Tests 22 VLMs on Agentic Tasks: The Best Scores Just 58.5%

A new offline benchmark of 1000 examples across five domains finds the best of 22 vision-language models answers correctly only 58.5% of the time. The main problem…

Cola2 beats Llama-3-8B on HellaSwag with 43 times fewer tokens

Hierarchical model Cola2 compresses text into continuous latents and scores 60.10 across eight tasks after training on 346B tokens, over 43x less than Llama-3's budget.

Verus-Insight Teaches an AI Video Detector to Find Artifacts and Prove Its Verdicts

AnyGameRenderer Dresses Gray AI Prototypes in Real Graphics and Keeps 62.9% of Units Intact

Business·Anthropic Research

Claude Proves Safety of 123 Libraries at the RISC-V Machine-Code Level

WeChat Open-Sources WeVisOne-VL: Detection, Masks and Matting in a Single VLM

Business·Anthropic Blog

Anthropic opens a call for life sciences researchers: 8–12 weeks with Claude, free of charge

Anthropic Hands Agent Skills Standard to AAIF: 1 Million Skills, 280 Million Installs

Cohere launches attested confidential inference on H100 GPUs in Azure

Research

Cohere Labs Tested AI With Riddles: 100+ Authors From 50 Countries Found Cultural Gaps

Business·Windsurf Blog

Cognition releases agent plugin to fix PHP and Node.js holes, including a CVSS 9.2 flaw

After the critical PHP2Shell flaws and a Node.js release fixing 10 CVEs, Cognition open-sourced a plugin that has Devin and other coding agents find and patch…

DirCast directs and commentates esports on its own, beating GPT-5

A single streaming video-language model picks the camera and talks for the caster in real time. DirCast's commentary beats GPT-4o in 29.2% of comparisons.

UserArena: A 64 729-Session Benchmark Tests Whether Shoppers Can Be Simulated

Researchers collected 64 729 real shopping sessions from 11 010 users and proposed OPTD, a method that reproduces the effect from a live A/B test more closely than…

Echo Teaches VLA Robots to Recognize Success and Failure Inside Their Own Actions

Researchers embedded stop and failure signals in every action chunk of a VLA model and denoise them faster than the motions. No per-frame progress labels are needed.

HiSTEVE 1.0 Plays Minecraft With a Mouse and Keyboard: 85.0% on MineStudio Bench