Der KI-Newsletter

Anthropic verhandelt den Kauf von Decart für 6 Mrd. Euro

Anthropic, gegründet von ehemaligen OpenAI-Wissenschaftlern, strebt darauf ab, das Unternehmen Decart zu übernehmen, das an großen Sprachmodellen arbeitet.

YouTube löscht 130.000 Kanäle wegen künstlicher Inhalte

Im Rahmen eines Crackdowns auf AI-generierte Inhalte löscht YouTube über 130.000 Kanäle.

KIMI K3-Sprachmodell schlägt top-geheimnisvolle AI-Vorhersagemodels

Im AI-Arena-Spielbezirk setzte sich das offen publizierte Sprachmodell KIMI K3 gegen die geheimgehaltenen Modelle Claude und GPT 5.6 durch.

SWEs Integrate AI in Development Workflow

Entwickler nutzen immer mehr KI-Tools wie Claude, aber dies birgt Sicherheitsrisiken und stellt menschliche Kreativität in Frage.

Dutch Buchhändler entdeckt: AI zerstört seltene Bücher

Ein niederländischer Buchhändler stellte fest, dass die Nachfrage nach seltenen Büchern als Spam oder Phishing betrachtet werden kann. Tatsächlich zerstören und vernichten AI-Unternehmen jedoch wertvolle physische Bücher für ihre Modelle.

Buch-Schreddern: KI-Firmen zerstören seltene Bücher für Daten

KI-Unternehmen schreddern seltene Bücher für Trainingsdaten – ein Skandal, der 800 Kommentare ausgelöst hat.

KI-Labore schreddern Millionen seltener Bücher – Sammler entsetzt

KI-Labore kaufen, scannen und schreddern Millionen seltener Bücher – Sammler schlagen Alarm.

AI Companies Are Shredding Rare Books to Feed Their Models

AI companies are destroying rare books to train models, sparking outrage and a debate about the ethics of data sourcing.

Google's Chief Scientist Quits to Launch AI Startup

OpenAI slashes GPT-5.6 prices as cost-sensitive enterprises push back, while former CTO releases a frontier model that Altman wouldn't.

AI Labs Shredding Millions of Rare Books to Feed Models

AI labs are buying, scanning, and shredding millions of rare, out-of-print books to train models, destroying irreplaceable cultural artifacts in the process.

AI Firms Shredding Rare Books to Feed Data Hungers

AI companies are literally shredding rare physical books to train models, and researchers are racing to digitize them before they're destroyed.

AI companies destroy rare and non-recoverable physical books

AI companies are destroying rare and non-recoverable physical books to train models, sparking outrage with 880 engagements.

AI labs shred rare books after scanning, sparking outrage

AI labs are buying and destroying irreplaceable rare books after scanning them, sparking outrage over cultural vandalism.

Mira Murillo's Frontier Model Dares OpenAI to Catch Up

Former OpenAI CTO Mira Murillo released her own frontier AI model, directly challenging OpenAI's dominance with a 'what Altman won't do' move.

AI firms pulping rare books to feed models, outrage grows

AI companies are destroying rare, non-recoverable physical books to train models, sparking outrage.

AI firms pulping rare books to train models, Dutch seller finds

OpenAI released GPT-transcribe and GPT-live-transcribe, specialized real-time transcription models built on GPT-5.6 architecture.

AI automation is eating agencies

OpenAI Adjusts Pricing for GPT-5.6 Models, Google Unveils Gemini Robotics ER 2, Zapier Introduces AI Agents, and Anthropic Launches Cowork Agent

OpenAI reduced prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%

Meta Starts In-House AI Chip Production

Meta begins in-house AI chip production in September 2026, signaling a major shift away from Nvidia dependency with $125-145B in AI capex.

OpenAI Drops GPT-5.6 Family with Tiered Pricing

OpenAI drops GPT-5.6 family (Sol, Terra, Luna) with tiered pricing and a new record on Terminal-Bench, plus a full-duplex voice model called GPT-Live.

OpenAI GPT-5.6: Three Tiers for Every Budget

OpenAI launched GPT-5.6 with three tiers (Sol, Terra, Luna), with Sol scoring just below Anthropic's Fable 5 at one-third the cost per task.

OpenAI releases GPT-5 Turbo: faster, cheaper, same smarts

OpenAI launches GPT-5 Turbo with 40% lower latency and 20% cheaper tokens, keeping reasoning quality on par with GPT-5.

Claude Fable 5 Now Costs Extra on Top of Subscription

Anthropic moved Claude Fable 5 behind a usage credit paywall, ending its inclusion in subscription plans and charging $10/$50 per million tokens.

OpenAI GPT-5.6 Restricted by US Government

Anthropic signed a $19B, 20-year infrastructure lease with TeraWulf, one of the largest dedicated AI compute deals to date, signaling a massive bet on physical capacity.

Claude Sonnet 5 Goes Agentic at Lower Cost

Anthropic's Claude Sonnet 5 launches as a cheaper, agentic mid-tier model that rivals flagship Opus 4.8 on reasoning and coding, with introductory pricing through August.

Claude 4 Opus vs Gemini 2.5 Ultra: Context War

Anthropic's Claude 4 Opus launches with a 2M token context window and a new Computer Use 2.0 API, while Google counters with Gemini 2.5 Ultra at 3M tokens, setting off a context-length arms race.

GPT-5.6 Restricted by US Government

US government preemptively restricts OpenAI's GPT-5.6 launch to approved partners, setting a major precedent for frontier model oversight.

Anthropic Ships Sonnet 5, Restores Fable 5

Anthropic drops Claude Sonnet 5 as default, restores Fable 5 globally, and pushes background agents into Claude Code.

OpenAI Drops GPT-5 Turbo with 2M Context

OpenAI released GPT-5 Turbo with a 2M context window and 50% lower costs, making it the go-to for long-document and agent workflows.

Mistral-Reason-7B: open-source reasoning model beats Llama-3-8B

Mistral releases Mistral-Reason-7B, an open-source reasoning model that outperforms Llama-3-8B by 12% on math benchmarks and runs on consumer GPUs.

OpenAI Gates GPT-5.6 Behind Government Preview

OpenAI gates GPT-5.6 behind government preview, with Sol hitting 88.8% on Terminal-Bench 2.1 and introducing ultra mode with subagents.

Enterprise AI Agent ROI Hits 171% Average

Enterprise AI agent ROI hits 171% on average, with customer service use cases delivering 41% in year one and scaling to 124% by year three.

OpenAI GPT-5.6 Launches with Government Gate

OpenAI's GPT-5.6 Sol, Terra, and Luna launch with government-gated access, marking the first official US frontier model restriction to ~20 approved organizations.

US Government Pulls Anthropic's Frontier Models

OpenAI and Broadcom unveiled their custom inference chip Jalapeño, delivering 50% cost savings and matching Nvidia Blackwell performance, with deployment expected by end of 2026.

OpenAI unveils Jalapeño AI inference chip

OpenAI shipped its custom Jalapeño inference chip with Broadcom for gigawatt-scale LLM serving starting late 2026.

GPT-5 Drops with Real-Time Video

OpenAI released GPT-5 with real-time video and audio, cutting voice latency to under 200ms and improving reasoning by 30% over GPT-4 Turbo.

AWS Blocks Lets Agents Build Their Own Backends

Enterprise AI is moving from chatbots to agentic workflows, with AWS, EDB, and Golem releasing tools that let agents operate on live data and infrastructure without human hand-holding.

SpaceX Signs $6.3B Compute Deal with Reflection AI

SpaceX signed a $6.3B compute lease with Reflection AI for Nvidia GB300 chips, massively expanding GPU availability and likely lowering costs for builders.

SpaceX Buys Cursor for $60B

SpaceX acquired Cursor (Anysphere) for $60 billion in an all-stock deal, signaling a major push to integrate xAI's models into enterprise coding tools.

SpaceX buys Cursor for $60B

SpaceX acquires Cursor (Anysphere) for $60B, signaling a major shift in AI coding tools and enterprise AI strategy.

SpaceX Acquires Cursor AI

SpaceX acquired Cursor to bring AI coding capabilities in-house, signaling a major shift in how tech giants are absorbing AI tools.

Claude Fable 5 returns after forced shutdown

Anthropic restored Claude Fable 5 and Mythos 5 after a six-day government-mandated shutdown, with the models still holding top benchmark records.

SpaceX acquires Cursor AI for $60B

SpaceX acquired Cursor AI for $60B in an all-stock deal, signaling AI coding tools are now critical infrastructure for hardware and aerospace engineering.

SpaceX acquires Cursor for $60B

SpaceX acquires Cursor (Anysphere) for $60 billion in an all-stock deal, signaling a major push to dominate AI coding tools with Grok integration.

Microsoft launches own AI models and M365 API

Microsoft launched its own AI model family and a GA API for M365 data, directly competing with OpenAI and giving developers a new way to build agents on enterprise data.

Anthropic deprecates Claude 4, forces migration today

Anthropic deprecated Claude 4 models today, forcing immediate migration to new versions and changing billing for agent SDK usage to separate credit pools.

OpenAI GPT-5 Turbo: 2M Tokens, Lower Cost

OpenAI dropped GPT-5 Turbo with a 2M token context window and 50% lower latency, undercutting Anthropic on price.

Apple Bets on Gemini to Power New Siri AI

Apple is ditching in-house Siri smarts and betting on Google's Gemini to power its new Siri AI platform, launching as a standalone app in iOS 27 and macOS 27 Golden Gate.

Anthropic Files $965B IPO, Launches Claude Managed Agents

Anthropic filed for a $965B IPO and launched Claude Managed Agents for production-grade scheduled workflows, signaling a massive validation of commercial frontier AI.

Claude Fable 5: Cheaper, Smarter, Safer

Anthropic drops Claude Fable 5 at half the price of previous frontier models, with 1M token context and a safety-routed fallback to Opus 4.8.

Claude Fable 5 crushes benchmarks at half price

Anthropic released Claude Fable 5 at half the cost of Mythos Preview, scoring 80.3% on SWE-Bench Pro and dominating GPT-5.5 and Opus 4.8.

Apple Rebuilds Siri with Gemini at WWDC

Apple is rebuilding Siri as a standalone app with a 1.2-trillion-parameter model built on Google Gemini, unveiled at WWDC today.

Anthropic Files for Trillion-Dollar IPO, Warns on AI Control

Anthropic filed for a trillion-dollar IPO while warning that Claude now writes 80% of its own code, signaling a critical juncture for AI control.

OpenAI GPT-5.6 previews agentic reasoning

OpenAI previews GPT-5.6 focused on multi-step agentic reasoning, retiring GPT-4.5 on June 27.

Google Gemma 4 12B Runs AI Locally on 16GB VRAM

Google launched Gemma 4 12B, a multimodal model that fits in 16GB VRAM, making local AI on Mac and mobile a real option.

Trump AI Executive Order Splits Industry

Trump's AI Executive Order creates a voluntary model sharing framework with the government, splitting tech leaders and sending AI stocks up on regulatory clarity.

Trump Signs Sweeping AI Executive Order

Trump signed an AI Executive Order requiring companies to share new models with the government 30 days before public release, shifting from laissez-faire to oversight.

Claude Opus 4.8 Ships Dynamic Multi-Agent Workflows

Anthropic launched Claude Opus 4.8 with dynamic multi-agent workflows, making it the default for Max, Team Premium, and Enterprise tiers.

Anthropic Hits $965B Valuation, Cuts Opus 4.8 Prices

Anthropic raised $65B at a $965B valuation, surpassing OpenAI, and cut Claude Opus 4.8 Fast Mode pricing by 3x to $10/$50 per million tokens.

Claude Opus 4.8 Drops Dynamic Agent Workflows

Anthropic releases Claude Opus 4.8 with Dynamic Workflows for parallel subagents, available on all three major clouds.

Claude Opus 4.8 Launches with Dynamic Workflows

Anthropic released Claude Opus 4.8 with dynamic multi-agent workflows and effort control, making it 4x less likely to hallucinate or hide code flaws.

Google and OpenAI employees back Anthropic in Pentagon fight

Workers Unite Against Military AI

OpenAI drops new fine-tuning API for code models

Google upgrades Search with Gemini 3.5 Flash AI Mode