🧠 Large Language Models

News about foundation models, LLMs, multimodal models, benchmarks, and model releases from OpenAI, Anthropic, Google, Meta, Mistral, and others.

1,576 articles

Qwen3 models target diverse reasoning, coding and everyday AI tasks

The Qwen3 series reflects a broader trend of AI models growing more capable at complex reasoning, coding, and everyday applications. Different versions are tailored to different use cases and performance needs. Source: trendhunter.com

Co-founder warns companies can't remove data once it trains an LLM

A co-founder of super{set} argues that businesses face a data dilemma similar to publishers who once handed content to Google — once information trains a large language model, it can't simply be extracted or audited afterward. The comment raises questions about data control and accountability in AI training. Source: ppc.land

ByteDance reportedly building a 10-trillion-parameter model, topping Anthropic's Mythos

Chinese tech giant ByteDance is developing a new large language model said to feature 10 trillion parameters, which would make it larger than Anthropic's flagship model. Details on training data and release plans remain unclear. Source: mezha.ua

AMD acquires Taalas to bake AI models directly into chips, sidestepping GPU memory limits

AMD's acquisition of Taalas targets the memory bottleneck that slows down GPU-based AI inference. Taalas' approach permanently encodes model weights into transistors, removing the need for DRAM during inference. Source: techtimes.com

Allstate rolls out its own LLM called ALLIE

Insurance giant Allstate has launched a proprietary large language model named ALLIE. CEO Tom Wilson said the company follows a 'technology-driven strategy' rather than one merely supported by technology. Source: insurancejournal.com

Tencent takes its open-source Hy3 AI model global

Tencent is expanding international availability of Hy3, its open-source large language model. The model is built for reasoning, coding, instruction-following and AI-agent tasks. Source: gulfbusiness.com

University of Alberta researcher builds local AI tool that keeps data off the cloud

Jonathan Schaeffer, professor emeritus at the University of Alberta, has developed a local AI platform aimed at researchers, writers and legal professionals wary of sending sensitive data to cloud services. The tool runs AI processing directly on users' own machines. Source: govtech.com

Anthropic model reportedly finds simple counterexample to 87-year-old math conjecture

A mathematician at Anthropic claims an AI model, referred to as Claude Fable 5, helped him identify a surprisingly simple counterexample challenging the long-standing Jacobian conjecture. The claim, if verified, would mark a notable case of AI contributing to pure mathematics research. Source: sciencedaily.com

UniSpec speeds up LLMs without retraining, works across hardware and languages

UniSpec is a new approach that accelerates large language models without any loss in output quality and without the need for retraining. It automatically adapts to different hardware setups and handles multilingual workloads with ease. Source: eurekalert.org

New hardware-aware framework speeds up LLMs without retraining

As large language models power more chatbots, virtual assistants, translators and coding tools, researchers have developed a framework that boosts their speed by adapting to the underlying hardware. Crucially, it requires no additional training to achieve these gains. Source: techxplore.com

LLM market projected to surge from $12.8B to $148.8B within a decade

The global large language model market is expected to reach USD 12.8 billion in 2026 and grow at a compound annual rate of 27.8%. Analysts forecast it will hit USD 148.8 billion by 2036. Source: futuremarketinsights.com

Google's first Southeast Asia AI report signals shift for Gemini and travel industry

Google published its inaugural Gemini Report: Southeast Asia 2026 on July 14, timed with the local-language launch of its new Gemini Spark agent. The report highlights how localized AI tools could reshape travel planning and services across the region. Source: webintravel.com

Study: Trust in AI medical advice depends heavily on user's own expertise

New research shows that people without medical training tend to blindly trust LLM-based diagnostic tools, even when the AI gives incorrect answers. Trained clinicians, by contrast, were much better at spotting and correcting the AI's mistakes. Source: goodmenproject.com

Anthropic researcher says Claude found a counterexample to an 87-year-old math conjecture

A mathematician at Anthropic claims he used the Claude Fable 5 model to find a surprisingly simple counterexample related to the Jacobian conjecture. The finding suggests advanced AI models could help mathematicians tackle open problems that have resisted proof for decades. Source: sciencedaily.com

Amazon brings Alexa+ to Australia, claims it beats rival chatbots

Amazon has launched its LLM-powered Alexa+ assistant in Australia and says it took extra steps to avoid the frequent errors seen in competing chatbots. The company positions accuracy and reliability as its key differentiator in the crowded AI assistant market. Source: afr.com

Why is Anthropic physically destroying books for AI training?

Commentator Kathryn James criticizes Anthropic for reportedly scanning and destroying physical books en masse to build training data, rather than negotiating copyright licenses. She argues the company chose the destructive route as a shortcut around legal complications. Source: theguardian.com

Study tests how well an LLM can triage children in the ER

Researchers evaluated a large language model's ability to assess urgency levels in pediatric emergency department triage, an area where human performance already varies widely between hospitals. The study aims to see whether AI could bring more consistency to these critical early decisions. Source: nature.com

The AI model you test isn't the one that actually ships

Most large language models used by the public are quantized versions, compressed after full-precision training to run more cheaply. The article warns that safety and behavior audits performed on the original model may not fully apply to the compressed version people actually use. Source: techpolicy.press

How to spot AI-generated text without relying on a detection tool

The piece pokes fun at self-proclaimed experts who confidently label online writing as AI-generated, often incorrectly. It argues that reliably identifying LLM-written content by eye is far trickier than most people assume. Source: towardsdatascience.com

DeepSeek quietly rolls out V4 Flash with two standout new features

DeepSeek has launched V4 Flash, a high-performance AI model offering very low pricing, alongside plans for a new suite of agent tools. The update is aimed at supporting developers and enterprise AI use cases. Source: eu.36kr.com