AI & ML

Mistral OCR 4 Adds Structure to Document AI: Bounding Boxes, Confidence Scores and 170 Languages

Mistral released OCR 4 on 23 June 2026, a document-intelligence model that returns bounding boxes, typed-block classification and inline confidence scores across 170 languages, and can run self-hosted in a single container. Mistral reports strong benchmark results but cautions that the scores are directional and the competitor numbers are its own reproductions.

2 Jul 2026 · 6 min read

AI & ML

US Export-Control Directive Leads Anthropic to Pull Fable 5 and Mythos 5 Worldwide

Anthropic disabled Fable 5 and Mythos 5 worldwide after a US Commerce Department directive restricted foreign access. Axios and other outlets report that warnings from Amazon and others preceded the order, but the motive and the severity of the risk remain contested. Anthropic calls the response disproportionate.

16 Jun 2026 · 8 min read

AI & ML

Anthropic Releases Claude Fable 5, Its First Public 'Mythos-Class' Model — With Guardrails and a Premium Price

Anthropic on 9 June 2026 released Claude Fable 5, the first generally available model from its 'Mythos' tier above Opus, alongside a more restricted Mythos 5. Anthropic reports large coding and knowledge-work gains, but Fable routes high-risk queries to Opus 4.8, costs about double Opus, and — per its own policy and independent reporting — ships with a 30-day data-retention requirement.

10 Jun 2026 · 7 min read

AI & ML

WWDC 2026: Apple Unveils 'Siri AI' and the Next Generation of Apple Intelligence

At WWDC 2026, Apple introduced 'Siri AI', an entirely rebuilt assistant, alongside a new generation of Apple Intelligence features — the long-delayed Siri overhaul it first promised in 2024, shipping as a free update this fall.

9 Jun 2026 · 6 min read

AI & ML

OpenAI’s New ChatGPT Memory System Turns Personalisation Into Infrastructure

OpenAI’s June 2026 ChatGPT memory update makes personal context more automatic, current and scalable. The product gain is obvious — but so is the need for users to understand what memory is doing.

6 Jun 2026 · 5 min read

AI & ML

Cognition raises US$1 billion at a US$26 billion valuation as Devin clears US$492 million in revenue

The maker of the AI software engineer Devin has raised US$1 billion, valuing the company at US$26 billion after the money. Eight months ago it was worth a tenth of that. The number under the round that matters more: Devin is now on a US$492 million annual revenue run-rate, with enterprise use growing 50 per cent a month.

1 Jun 2026 · 3 min read

AI & ML

OpenAI Announces First Applied AI Lab Outside the US in Singapore, Backed by S$300 Million Commitment

OpenAI signed a memorandum of understanding with Singapore's MDDI on 20 May 2026, committing more than S$300 million and announcing the city-state as the location of its first Applied AI Lab outside the United States — with over 200 technical roles planned over the next few years.

1 Jun 2026 · 5 min read

AI & ML

NVIDIA Opens Singapore AI Research Lab as City-State Launches Singapore's First Multi-Operator Robotics Testbed, Planned for Later in 2026

Singapore announced two linked moves on 20 May: NVIDIA confirmed its first research lab in the city-state, and the government revealed Punggol Digital District will become Singapore's first multi-operator robotics testbed — with eight named industry partners and trials planned for later in 2026.

1 Jun 2026 · 5 min read

AI & ML

Alibaba's Qwen 3.7 Max Debuts as Top-Ranked Chinese AI Model, Built to Run Autonomously for Dozens of Hours

Alibaba's Qwen 3.7 Max placed fifth globally on the Artificial Analysis Intelligence Index at launch — the highest-ranked Chinese model — after a 35-hour, 1,158-tool-call autonomous demonstration at the Alibaba Cloud Summit. The model signals a shift from chat-leaderboard competition toward long-horizon agentic enterprise contracts.

1 Jun 2026 · 5 min read

AI & ML

NVIDIA Launches Nemotron 3 Ultra at Computex 2026 — 550B-Parameter Open-Weights Model Targets Enterprise Agentic AI

NVIDIA unveiled Nemotron 3 Ultra at Computex 2026 in Taipei — a 550B-parameter mixture-of-experts model with 55B active parameters, scoring 48 on Artificial Analysis's Intelligence Index and claiming 300-plus tokens per second. It leads US open-weights models but trails China's Kimi K2.6 at 54.

1 Jun 2026 · 5 min read