Latest news, developments and reporting on Large Language Models.
Amazon Web Services' Strands Labs released Strands Decider 2B, an open-source 2-billion-parameter decision model that selects among predefined options for agentic AI workflows. It is available on Hugging Face and GitHub.
Family offices made 42 direct investments in September, down from 52, while backing Mistral, Precision Neuroscience and geothermal startup Mazama, Fintrx data show.
Hugging Face released Olmo-core 3 on Oct. 1, 2026, an open training framework for large mixture-of-experts models. The system is designed to scale MoE training to more than a trillion parameters while reducing communication and memory costs, the company said.
Anthropic asks Australia to permit AI training on copyrighted works through conditional approval and an opt-out model, while the ABC and SBS demand AI firms face the same copyright, defamation and privacy rules as media outlets.
The 10-year Treasury yield touched a 24-year high before easing, while U.S. stock futures rose. Micron reported strong results but fell premarket, Google launched Gemini 4 Argon, and Broadcom agreed to lend Anthropic up to $42 billion.
Google has released Gemini 4 Argon, its most advanced model, which it says rivals OpenAI's GPT-6 Astra and Anthropic's Opus on key benchmarks while costing less per task, and can autonomously find and patch software vulnerabilities.
The Guardian has opened autumn submissions for its Young Country Diary, asking UK children aged 8 to 14 to write 200 to 250 words about a recent encounter with nature. The deadline is noon on 2 November.
Apple may launch a smart display, HomePod mini 2 and faster Apple TV 4K on Oct. 13, Android Authority says, citing Bloomberg.
NVIDIA released Kumo Tabular, open tabular foundation models for one-pass prediction with commercial-use weights.
Three consortia led by SK Telecom, KT and Kakao will launch closed beta AI services in October under South Korea's government-led AI for All project, Yonhap News reported.
Perplexity Research and turbopuffer have released pplx-embed-v2-context-9b-preview, a self-hosted contextual embedding model for RAG pipelines that is trained to retrieve answers and their supporting evidence.
Google launched Gemini 4 Argon, claiming top scores in 13 of 18 benchmarks, but access is limited amid internal doubts.
OpenAI says it disrupted a campaign to extract protected AI reasoning and links a core cluster to China's Moonshot AI.
Alphabet's Google released Gemini 4 Argon, which it calls its most powerful model, and is rolling it out first to a select group of cyber partners through its Fairwind security program.
Google has unveiled Gemini 4 Argon, a frontier AI model that it says leads or ties rivals on 13 of 18 disclosed benchmarks. It is initially limited to trusted cyber defenders and select testers, with broader availability planned later.
Google began rolling out Gemini 4 Argon, a frontier model it says leads Anthropic and OpenAI on most of its benchmarks, to vetted cybersecurity defenders, with general developer access still undated.
OpenAI released GPT-6.1 Sol at one-fifth of Astra's rates, citing near-Astra coding and computer-use benchmarks.
GMKtec launches EVO-X5 Pro at $6,399 with AMD Gorgon Halo chip, 192GB memory, and offline 320B-parameter LLM support.
Five technology CEOs and OpenAI President Greg Brockman signed a White House-backed voluntary AI safety accord that urges frontier model developers to adopt internal controls, external audits and board oversight.
Google announced Gemini 4 Argon, its most advanced AI model, initially available only to trusted cyber defenders.
OpenAI unveiled a Decisions API at Dev Day that mirrors TypeSafe's Jev, a fast, cheap classifier now being tested as a low-cost way to monitor and secure AI agents.
Google's pilot paying publishers for AI answer content is yielding tiny payouts, The Information reports.
Ajo announced Strata on Hacker News, an early-beta semantic layer that combines strict naming, semantic routing, complex measures and cohort filters. It offers dashboards, subscriptions and Google Sheets exports via agent conversation or MCP, with a free Docker tier for up to 25 users.
Kalshi traders increased the odds that Anthropic will announce an IPO this year after reports said the company plans to warn investors about dangerous model risks.