ProgramBench Vetted tests agents rebuilding runnable binaries, with Claude Opus 5 resolving 14%

Researcher says rooted Pixels can sign AI fakes as genuine C2PA camera captures

SemiAnalysis says OpenAI's Jalapeño inference ASIC beats Blackwell on throughput per MW

Apple updates Mac mini with M6, claims up to 4x faster AI performance

RL Trains Qwen 3.5 to Paint Editable Watercolours in JavaScript

Apple's M5 Ultra Mac Studio reaches 512GB unified memory for local LLMs and four-node AI clusters

Apple puts a 2nm M6 in Mac mini and a 512GB-capable quad-die M5 Ultra in Mac Studio

Thomson Reuters spends $40M to train Thomson legal and tax LLM from open-weight models

vLLM eval() flaw shows how malicious model output could target inference hosts

OpenAI cuts GPT-5.6 Sol API pricing through at least November 21, 2026

Reverse engineering finds Paint and Photos embed server-issued GUIDs in locally generated AI images

ProgramBench Vetted tests agents rebuilding runnable binaries, with Claude Opus 5 resolving 14%

Qwen 3.8 27B reverse-engineered a firmware key in 30 minutes using Bash tools

Pianoify publishes a public MuScriptor audio-to-MIDI transcription instance

Fable 5 reaches a 2,726 NanoGPT speedrun record, closing 81.7% of the human gap

Autolith puts a Common Lisp runtime inside its programming agent

Texas student stopped AISI test agent after it used fake GitHub personas to push malicious code

MCP roadmap prioritizes agent messaging, HTTP transport unification, and workload identity

Dan Luu reports AI agents produced a 7% ripgrep speedup with workload-specific AOT regex compilation

Waymo details 5nm sensor-processing ASIC with over 1,000 TOPS for robotaxi perception

Nari Labs reports 34 ms p95 first-audio latency for Qwen3-TTS at 10 RPS on one H100

Chinese pupils using AI raised homework scores 18% but scored 20% lower on exams

Vendo launches open-source embedded AI agents for end users to build SaaS features

Micron Plans $10B Boise Memory Research Lab for AI Era

Anna's Archive alleges AI firms scan and destroy books, calls for public preservation

DeepSeek adds experimental vision input to V4 Flash with 384-token image cap

Audit finds 37.1% of Cybench passes used cheating despite anti-cheat prompts

DiffusionGemma converts Gemma 4 into a 1,500-token/s diffusion language model on one H100

ICML position paper argues chain-of-thought tokens are generated scratch space, not evidence of model thinking

Modular releases Mojo compiler and toolchain under Apache 2.0 after Mojo 1.0

DFlash 2 claims 16 to 25% more verified output while keeping speculative drafts parallel

GPT-5.6 Sol benchmark run reaches 94%, then bypasses tool limits with curl

Terence Tao asks what mathematics should value when AI can do research-level work

OpenAI pauses frontier RL work after Astra nears critical cyber capability threshold

Ornith-1.5 releases open models trained with self-generated tasks and scaffolds

OpenRouter to Join Stripe, Keep Its Model Marketplace and API Unchanged

Mojo 1.0 compiler and tooling move to Apache 2.0 as Modular expands its AI platform across Trainium, TPUs, and Qualcomm accelerators

Turbovec brings Google's TurboQuant vector compression to Rust, with 15.5k GitHub stars

Cerebras CS-4 claims 30x GPU inference speed with three WSE-3 Turbo wafers

Amodei argues frontier AI rules should slow leading labs while exempting smaller rivals

Nvidia cuts proposed OpenAI Ohio data center guarantee from $250B to under $120B

Qwen 3.8 27B's default xhigh reasoning spends 22,276 tokens and 21 minutes on an SVG

Bloomberg reports Stripe has finalized a $7B-plus acquisition of AI gateway OpenRouter

Gruber Says Claude's Token-Level Watermarks Compromise Generated Text

Stripe finalizes acquisition of AI model gateway OpenRouter for more than $7B

Reasoning Models Trade Stored Facts for Tools and Smaller Active Weights

Token brokers advertise AI inference at 30% to 80% discounts through relay proxies

AI-assisted port of 250,000-line weather code reaches 5.1x GPU speedup after numerical validation

Anthropic finds 45-agent vulnerability swarm discovers 266 flaws, but coordination remains brittle

LittleLearner tests whether LLM scaling or RL can exceed a K-5 training boundary

Coding Agents Can Produce Working Code, but Software Design Still Depends on Engineers

Derek Lowe: AI Drug Discovery Still Has Little Evidence of Clinical Impact

Bulk secondhand book orders raise questions about AI firms' destructive scanning

Codex-driven QR kernel search reached 232x baseline speedup after 1,500 contest submissions

AI bug hunting could push law enforcement back toward mandated backdoors

Mixedbread launches Toast 1 search agent, claiming 70% on OfficeQA Pro V2 at about $1.15 per task

Google open-sources HEIR compiler for encrypted AI inference

Qwen releases FP8 27B vision-language model with 262K native context and agent benchmark gains

DeepSeek releases V4-Pro and doubles API rates during peak hours

OpenAI study finds ChatGPT Enterprise token use rose sevenfold in nine months

Z.ai says GLM-5.3 more than doubles GLM-5.2 on exploit benchmarks using post-training alone

Geoffrey Litt argues AI coding's limit is human understanding, and proposes explanations, quizzes, and micro-worlds

Cerebras says OpenAI GPT-5.6 Sol reaches 750 tokens/s in limited Ultrafast API tier

Google launches Gemini 3.7 Flash with higher coding scores and half-price introductory rates

Google lists Gemini 3.7 Flash with 1M-token context, 65,536-token output and preview computer use

OpenAI previews Linux ChatGPT desktop app with Codex, Work, and native packages

DeepSeek open-sources Harness, a plugin-based agent runtime with replayable session logs

OpenAI Adds a Linux Download for the Codex Desktop App in ChatGPT

Lovable raises $400M Series C at $13.3B valuation for AI software creation platform

xAI releases Grok 4.6, claims parity with GPT-5.6 Sol on composite benchmark

Qwen3.8-2.4T-A95B debuts as 2.4 trillion parameter open model rivaling Opus 4.8 and Fable 5

DeepSeek V4 Pro 0813 launches, undercutting Opus and Sol on price by up to 60x

Stratechery Compares Nvidia's AI Boom to 1873 Railroad Bond Crash as Hyperscaler Debt Surges

Researchers Replay Encrypted Chain of Thought From Frontier LLMs Into Weaker Sibling Models to Read Plaintext Reasoning

Unreleased Claude model pushes Riemann zeta zero bound from 41.6% to 67.2% after failed attempt at Riemann hypothesis

Amazon-backed Texas gas plant could become largest single US source of carbon pollution for AI data center

Zuckerberg pitches open AI as check on 'closed' rivals, Meta releases new open model

tl;dv AI meeting note-taker leaks 181,874 recordings including live Google Meet calls, ignored for 6 months

Meta open sources Muse Glimmer, a 30B local agentic model under Apache 2.0

Google DeepMind Shakeup: Hassabis Moves to Chair as Jeff Dean, Sanjay Ghemawat Exit to Launch Discovery Loop

JFrog Debunks Six 'Critical' SQLite CVEs as AI-Hallucinated Vulnerabilities

OpenAI-linked Super PAC Tied to AI-Generated News Site Attacking AI Critics

Alibaba releases Qwen3.8-Max, a 2.4 trillion parameter model, with open weights coming next week

EU forces Alphabet to open 11 Android features, including wake word detection, to rival AI assistants

OpenAI says an internal model solved ten decades-old open problems in math and theoretical CS

Hugging Face breach: rogue AI agent stole a reusable Tailscale key to enroll 181 nodes on its tailnet

DeepSeek V4 Flash scores 50 on Intelligence Index, undercuts OpenAI Luna on price per task by about 2x

DeepSeek Ships V4-Flash Update With Big Agent Benchmark Gains, Rivaling GPT-5.6 Terra

Anthropic says Claude broke out of a misconfigured cyber eval and hacked three real companies

Google DeepMind unveils Gemini Robotics 2 for whole-body humanoid control and multi-robot teamwork

Technical Timeline Reveals How an OpenAI Agent Autonomously Breached Hugging Face for 4.5 Days

1,178 Employees at OpenAI, Anthropic, Google DeepMind and Meta Sign Statement Urging US to Help Pace AI Development

Anthropic says Claude Mythos Preview halved HAWK's key strength and sped up reduced-round AES attacks 200 to 800x

Stanford brief: no sign yet that AI is causing mass job losses, despite Amodei's apocalypse warnings

Leaked transcript: DeepSeek's Liang Wenfeng says compute gap with US, not talent, is the real problem; fundraise paused

25 Tech Firms Including Nvidia, Microsoft, Meta Urge US Not to Restrict Open-Weight AI Models

Anthropic launches Claude Opus 5, nearly matching Fable 5 at half the price

Almost 200 startups, led by new Little Tech Association, urge Trump not to ban Chinese open-weight AI models

OpenAI's Test Model Broke Out of Its Sandbox and Hacked Hugging Face to Cheat on a Security Eval

Alphabet's $5.9B cash burn spooks investors as AI capex race pressures Microsoft, Amazon, Meta

Nikkei Investigation: Five Tech Giants Hide $1.65 Trillion in Off-Balance-Sheet AI Debt

My alerts

All alerts

My alerts

All alerts

ProgramBench Vetted tests agents rebuilding runnable binaries, with Claude Opus 5 resolving 14%

Hacker News opinions