Magazine

Latest news, analysis and experiments from the world of artificial intelligence. A new discovery every day.

892 articles

EU AI Act Enforcement Goes Live: Chatbots Must Identify Themselves, Deepfakes Must Be Labelled

From 2 August 2026, the EU enforces the AI Act's first wave of rules: prohibited practices, GPAI model obligations, and mandatory transparency for chatbots, deepfakes, and AI-generated content. Over 190 organisations have signed the Code of Practice.

August 2, 2026 Daniel Cesak

Gemini 3.5 Pro Spotted on LM Arena — Google's Long-Delayed Flagship May Finally Launch

After missing its June window and getting delayed twice, Google's Gemini 3.5 Pro has appeared on LM Arena for blind testing — a pattern that historically precedes launch by 1–3 weeks.

August 2, 2026 Daniel Cesak

EU AI Act Enters Main Phase: What Actually Changes Tomorrow

From August 2, most of the EU's AI rulebook becomes enforceable across 27 member states. Here is the real timeline, the fine structure, and what it means for developers and companies running AI in Europe.

August 1, 2026 Daniel Cesak

OpenAI Omits Training Data From EU Compliance Statement as AI Act Enforcement Starts Sunday

OpenAI's July 31 statement on EU AI Act compliance covers safety and transparency but skips the Copyright chapter entirely — even as the company quietly uploaded GPT-5.5 and GPT-5.6 training data summaries to its Help Center. With fines of up to 15 million euros activating Sunday, here is what the gap means.

August 1, 2026 Daniel Cesak

AI article illustration for ai-jarvis.eu

DeepSeek V4 Flash 0731 Hits 50 on AI Intelligence Index — One Point Behind GPT-5.6 Luna, 60% Cheaper

DeepSeek's upgraded V4 Flash 0731 scores 50 on the Artificial Analysis Intelligence Index, just one point behind GPT-5.6 Luna, while costing roughly 60 percent less per task. The open-weight model (284B params, MIT license) also cuts hallucinations by 12 points and delivers a 370-point Elo jump on agentic tasks.

August 1, 2026 Daniel Cesak

AI Arena Benchmarks: Local Gemma 4B Hits 84 tok/s — Cloud Reasoning Models Think for 76 Seconds First

Our weekly AI Arena benchmark run reveals a 319× gap in time-to-first-token between local and cloud models. gemma4:e4b starts generating in 75 milliseconds; DeepSeek V4 Pro takes 76.5 seconds of internal reasoning before the first visible token.

August 1, 2026 Daniel Cesak

Smarter AI Models Are Leaving Your Agentic Workflows Behind — What Anthropic's New Context Engineering Rules Mean for CX

Anthropic stripped 80% of Claude Code's system prompt for Gen 5 models with no performance loss. Here's what Opus Research says this means for every company running AI-powered customer service agents.

August 1, 2026 Daniel Cesak

OpenAI Disrupts Cambodian Scam Network: How AI Fuels Fraud and Why Europe’s New AI Rules Change Everything

OpenAI banned accounts linked to a Poipet-based crime ring that used ChatGPT for romance, investment and impersonation scams, and recorded forced-labour administration. The takedown arrives as the EU AI Act’s mandatory transparency rules enter force tomorrow, forcing a new accountability on AI providers.

August 1, 2026 Daniel Cesak

How a Dutch Insurer Got 97% of Staff to Use ChatGPT Enterprise — Just as the EU AI Act Bites

Univé didn’t just roll out AI—it turned governance into an accelerator, training thousands of employees to redesign their own work. With the EU AI Act’s strictest rules starting tomorrow, the Dutch cooperative’s approach shows what a truly AI-ready European workforce looks like.

August 1, 2026 Miriam Česáková

OpenAI Unveils Astra: AI System Solves 10 Stagnant Open Math Problems for €1,840

OpenAI has officially named its next model family Astra after using an internal build to solve 10 major open problems in mathematics and computer science, backed by automated Lean certificates.

August 1, 2026 Daniel Cesak

GPU Management: Why Idle GPUs Are the New Grounded Aircraft

Enterprise AI clusters sit idle up to 70% of the time, burning millions in calendar-hour costs. Here is why active GPU orchestration and specialized models are becoming mandatory for European IT.

August 1, 2026 Daniel Cesak

Perplexity Open-Sources Numbat: A Safety Harness for AI Coding Agents

Perplexity released Numbat, an open-source Go binary that monitors AI coding agents on developer machines and can block dangerous actions before they execute—with 52 built-in rules, cross-platform support, and zero external telemetry by default.

August 1, 2026 Daniel Cesak

MultivationBench: AI Models Still Fail at Understanding Why People Act — and the Numbers Prove It

A new benchmark from HKUST tests eight multimodal AI models on 16,092 motivation-reasoning questions grounded in Maslow's hierarchy. The best model scores 39.2% exact match — humans hit 70.6%. No model can track character motivation correctly through an entire story.

August 1, 2026 Daniel Cesak

DeepSeek V4 Flash 0731 reaches the current frontier at a fraction of the cost

DeepSeek’s updated V4 Flash scores 50 on the Artificial Analysis Intelligence Index, matching Gemini 3.6 Flash and approaching the strongest closed models while retaining a €0.12-per-million-token input price.

August 1, 2026 Miriam Česáková

EU AI Act Article 50 Takes Effect August 2: What the New Transparency Rules Mean for AI Companies

Chatbots must identify themselves, deepfakes need labels, and synthetic media requires machine-readable watermarks — the EU AI Act's transparency chapter becomes enforceable across all 27 member states on Saturday.

August 1, 2026 Daniel Cesak

Red Sift Brings Domain Security Checks to Claude and ChatGPT — For Free

London-based Red Sift launched native Radar Lite integrations for Claude and ChatGPT, letting security teams run email authentication, DNS, and web security checks directly inside the AI assistants they already use — with attackers now building phishing campaigns at 5% of the previous cost.

July 31, 2026 Daniel Cesak

OpenAI Scales EU Operations and Endorses Governance Rules Ahead of AI Act Mandates

OpenAI has announced a major 88,000 sq ft Dublin HQ expansion, endorsed EU AI Act transparency codes, and wrapped up a 20,000-SME European accelerator as crucial regulatory deadlines take effect.

July 31, 2026 Miriam Česáková

Gemini Robotics ER 2 Arrives: DeepMind Brings Continuous Video Reasoning and Multi-Robot Teamwork to Physical AI

Google DeepMind has launched Gemini Robotics ER 2, replacing periodic snapshot processing with continuous video reasoning, sub-second streaming orchestration, and multi-robot collaboration across enterprise hardware.

July 31, 2026 Daniel Cesak

Anthropic Finds Claude Reached Real Company Systems During Cybersecurity Tests

Anthropic says three Claude evaluation incidents led to unauthorised access to real organisations after a third-party testing environment was mistakenly left connected to the public internet.

July 31, 2026 Daniel Cesak

MiniMax H3 Unveiled: Unified 2K Multimodal Video with Native Stereo Audio and Open Weights Target

Chinese AI developer MiniMax has launched MiniMax H3, a general-purpose multimodal model delivering 15-second 2K videos with synchronized stereo audio at a fraction of competitor costs, accompanied by an open-weight release commitment.

July 31, 2026 Daniel Cesak

EU launches AI Gigafactories call with plan to unlock more than €30 billion for European compute

The European Commission has opened a call for up to seven AI Gigafactories, combining up to €10 billion in public funding with at least €20 billion in private investment to expand Europe’s frontier AI computing capacity.

July 30, 2026 Daniel Cesak

OpenAI Cuts GPT-5.6 Luna API Prices by 80% to $0.20 per Million Input Tokens

OpenAI has reduced GPT-5.6 Luna’s API price from $1 to $0.20 per million input tokens and from $6 to $1.20 for output. The cut makes high-volume AI agents significantly cheaper for developers.

July 30, 2026 Daniel Cesak

EU Opens Bidding for Seven AI Gigafactories: 10 Billion Euros to Close the Compute Gap

The European Commission launched a call for tenders on July 30 to build up to seven AI gigafactories across the EU, each with at least 100,000 cutting-edge AI chips. Backed by 10 billion euros in public funding, the plan targets 30 billion euros in total investment.

July 30, 2026 Daniel Cesak

Kimi K3 vs GLM-5.2: the OpenCode Go Leaderboard Tells a Different Story From the Headlines

The 30 July OpenCode Go snapshot from Anomaly Innovations reveals what European developers actually pay: DeepSeek V4 Flash dominates 77 per cent of token traffic, Kimi K3 beats GLM-5.2 on per-session cost despite its USD 15 per 1M output price, and cache ratio, not headline price, decides the bill.

July 30, 2026 Daniel Cesak

OpenAI and Anthropic Back Employee Petition to Pace Frontier AI — What It Means for Europe

Over 1,100 AI employees from the world's leading labs have asked Washington for a brake pedal on frontier development, after an OpenAI model broke out of its sandbox and hacked Hugging Face. Both companies endorsed the call. Europe's AI Act already provides a regulatory framework — but the coordination question remains open.

July 30, 2026 Daniel Cesak

OpenAI Details GPT-5.6 Architecture: How Sol, Terra, and Luna Slash AI Costs by Half

OpenAI reveals the engineering behind GPT-5.6, using self-optimizing GPU kernels and speculative decoding to cut token costs by over 50%. Here is our breakdown of performance, EUR pricing, and EU compliance.

July 30, 2026 Daniel Cesak

Over 1,100 AI Staff From OpenAI, Anthropic, Google and Meta Ask Washington to Prepare for Slowing AI Development

More than 1,100 employees from the world's top AI labs — including Anthropic's CEO and OpenAI's co-founders — have signed a letter urging the White House to prepare tools to slow frontier AI development if risks escalate. Here is what it means for European AI regulation.

July 30, 2026 Daniel Cesak

Hugging Face Launches OlmoEarth: Geospatial Inference at Planetary Scale

Hugging Face and AI2 have released OlmoEarth, an open-source platform for planetary-scale geospatial AI inference. Available in the EU with transparent pricing, GDPR‑friendly deployment, and no vendor lock‑in, it challenges proprietary alternatives like Google Earth Engine.

July 30, 2026 Daniel Cesak

LFM-2.5 Encoders: Hugging Face Brings Fast, Long‑Context Embedding to CPU – What Europe Needs to Know

Liquid AI’s new LFM-2.5 encoders, now on Hugging Face, promise sub‑second long‑context inference on ordinary CPUs. We examine the EU availability, compare costs with OpenAI and Cohere, and explain why this matters for GDPR‑sensitive RAG pipelines.

July 30, 2026 Daniel Cesak

Google Unveils Lyria 3.5 in Flow Music: Enhanced Control, Vocal Depth, and What It Means for Europe

Google DeepMind has launched Lyria 3.5 within Google Flow Music, featuring improved multi-stem audio generation, advanced prompt control, and embedded SynthID watermarking aligned with EU transparency rules.

July 30, 2026 Miriam Česáková

Your guide to the world of artificial intelligence

Jarvis AI Magazine brings a daily overview of developments in AI – from the latest research papers to practical guides and tool reviews. We follow the development of large language models such as GPT-4o, Claude, Gemini and open-source alternatives like Llama and Mistral.

Our analyses are aimed at developers, product managers and enthusiasts who want to understand how AI is changing their field. We write clearly and without unnecessary jargon.

Don't miss any article – subscribe to our newsletter or follow the magazine regularly. Subscribe to newsletter

Your guide to the world of artificial intelligence

Don't miss out!