APREX · EST. 2026
Underground
I FOKUS
Microsoft Gathers Developers in SF: New AI Models and Windows Moves at Build 2026
Microsoft is holding its Build conference in a smaller, more intimate format in San Francisco this week, with new AI models and Windows improvements on the agenda — just as developer trust is at a historic low.
Flere saker.
6 DAYSPolicyTrump reverses course: Lifts export controls on Anthropic's most powerful AI modelsJUL 1ModelsClaude Sonnet 5 cuts prices and takes on GPT-5.5 on agentic tasksJUN 28ToolsApple Gives Google Access to Siri — Experts Warn on PrivacyJUN 27PolicyWhite House Halts GPT-5.6: Demands Safety Clearance Before LaunchJUN 24ModelsOpenAI and Broadcom Unveil Jalapeño: 50% Cheaper Than Nvidia's BestJUN 23UndergroundClaude pretends to think — but doesn't show you the real thoughtJUN 23SecurityOpenAI Takes Aim as Anthropic's AI Breaks Into NSA SystemsJUN 20ResearchGPT-5.4 optimized key reaction in medicinal chemistry almost entirely on its ownJUN 18UndergroundWhich AI is controlling the robot running toward you?JUN 15SecurityChina Reportedly Gained Access to Anthropic's Secret AI ModelJUN 14ModelsAnthropic shut down Fable 5 for the entire world – warns of historic precedentJUN 10ModelsAnthropic releases Claude Fable 5 — most powerful model everJUN 9ModelsGoogle's new 12B model runs advanced AI directly on your laptopJUN 7ModelsChinese AI code crushes American prices. The quality? It's complicated.JUN 6ModelsClaude Opus 4.8 spawns hundreds of AI agents simultaneously. GPT-5.5 loses benchmarks.JUN 1ModelsNVIDIA's Cosmos 3 Merges the Brain and Body of Robots into a Single ModelMAY 29ModelsGemini Omni reveals Google's new video ambitionMAY 29ModelsLlama 4 Makes Multimodal AI Core Meta InfrastructureMAY 29ModelsDeepSeek-R1-0528 keeps the reasoning pressure onMAY 29ModelsGemini 3.5 Flash Makes Google's Agent the Default ModelMAY 29ModelsClaude Sonnet 4.6 Brings Opus-Level Power at a Lower PriceMAY 29ModelsGPT-5.2 Makes Office Work OpenAI's New BattlegroundMAY 29ResearchPaperBench tests whether AI can replicate real researchMAY 29Open SourceKimi K2 shows that open agentic models can now compete at the trillion-parameter levelMAY 29Open SourceLlama 4 Brings Open Multimodal AI to the Mass MarketMAY 29Open SourceGemma 3 Makes Multimodal AI Possible on a Single GPUMAY 29Open SourceOpenAI reopens the weights: gpt-oss brings local reasoning into the mainstreamMAY 29Open SourceQwen3-Coder turns open-source coding AI into a geopolitical developer battleMAY 29Open SourceCohere releases open enterprise model: Runs on just two H100sMAY 29SecurityOpenAI opens AI biodefense to authorities: GPT-Rosalind to stop pandemicsMAY 28UndergroundAnthropic releases Opus 4.8 — beats GPT-5.5 on 12 benchmarksMAY 20ModelsGemini Omni transforms images, audio, and text into video in one conversationMAY 6ModelsGPT-5.5 Instant: 52.5% Fewer Hallucinations in New ChatGPT StandardMAY 4UndergroundChinese Open-Source Model Beats Claude, GPT-5.5, and Gemini in CodingAPR 28ModelsNVIDIA's new AI model beats competitors with 3.3x higher performanceAPR 26ModelsOpenAI Releases GPT-5.5 and Chases the 'Super App' ThroneAPR 25SecurityAI gives hackers 192x faster attacks — security chief raises alarmAPR 25IndustryGoogle commits $40 billion to AnthropicAPR 24UndergroundDeepSeek-V4 Drops and Crushes Everything Open Source Has SeenAPR 23UndergroundAnthropic admits: Claude Code got worse — here's what went wrongAPR 22SecurityAnthropic's Most Dangerous AI Escapes Control – Unauthorized Users Gained AccessAPR 21IndustryAmazon Pumps $5 Billion into Anthropic – Gets $100 Billion Back in AWS PurchasesAPR 18UndergroundxAI Wipes Out Competitors with Grok Voice API — 5x Faster Than GPTAPR 17UndergroundAnthropic Released an 84-Page AI Soul — and HN is ExplodingAPR 17ModelsOpenAI's Codex takes control of your PC – without you lifting a fingerAPR 16SecurityOpenAI gives $10M to cyber defense — GPT-5.4-Cyber in battle against hackersAPR 15UndergroundAnthropic Releases Claude Code Routines — AI Agents That Code While You SleepAPR 14IndustryOpenAI Chief Warns Employees: Anthropic Stealing 40% of Enterprise MarketAPR 12SecurityAnthropic Gathers Apple, Google, and 45 Others to Stop AI HackingAPR 11ModelsAnthropic's Mythos Raises Alarms: Who Is Really Most Vulnerable?APR 10UndergroundClaude mixes up who says what — and it's a real problemAPR 9SecurityAnthropic's Secret AI Found Thousands of Security Flaws — Kept from PublicAPR 7UndergroundAnthropic's Secret AI Finds Bugs Older Than the Internet