🧠 Large Language Models

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

1,700 articles

iFlytek Open-Sources Spark-X2.6 LLM Built on Fully Domestic Chinese Computing

Chinese tech company iFlytek has released and open-sourced Spark-X2.6, a general-purpose large language model, at an industry event in Beijing. The model runs entirely on domestic computing power. Source: globaltimes.cn

Giving AI a 'Crystal Clear' View of the Materials World

In principle, AI could make choosing the right material for an application as easy as ordering dinner, with scientists simply telling a large model what they need. The article explores how to give AI a clearer picture of materials. Source: technology.org

Martingale Launches Raven Alpha for AI Model Drift Detection

Martingale, founded by three Carnegie Mellon computer science students, has released Raven Alpha, a model drift detection method. It ranks first in benchmark tests. Source: usatoday.com

VDR.ai and LLM.co Team Up on Custom AI Hardware and Private LLMs for Finance

The partnership combines AI-powered virtual data rooms, dedicated GPU infrastructure and private large language models. It targets investment banks and other financial institutions. Source: financialcontent.com

Build Custom NER Pipelines with Scikit-LLM

A tutorial shows how to assemble standardized, custom named entity recognition pipelines in a few simple steps. The approach is powered by the reasoning abilities of LLMs. Source: kdnuggets.com

Open-weight models could help Türkiye keep control of its AI and data

Open-weight AI models could help Türkiye protect sensitive data, lessen reliance on foreign providers and build lasting institutional intelligence, according to Daily Sabah. Source: dailysabah.com

The hidden cost of LLM routing: why switching models can cost more

A hands-on test of LLM routing shows that switching models in the middle of a session can wipe out the savings that cheaper inference promises. Source: towardsdatascience.com

From chat demos to production: testing LLM systems at scale with DeepEval

Building a generative AI chat prototype is now easy, requiring little more than an LLM, a vector database and a frontend. The real challenge is evaluating such systems reliably at scale, which is where DeepEval comes in. Source: blogs.worldbank.org

Specialized platform beats general LLMs in 475-variant clinical classification

General-purpose large language models were considerably less consistent than a specialized clinical interpretation platform when classifying genetic variants. The comparison covered 475 variants. Source: bioengineer.org

Autonomous 'Skynet' agent hijacks exposed LLM endpoints to power attacks

Attackers have created an autonomous offensive agent that scans the internet for exposed large language model (LLM) endpoints. It then takes them over and uses them to carry out its own attacks, a technique dubbed LLMjacking. Source: security.com

Hidden trigger words could turn LLM-powered robots dangerous; researchers build a defense

LLMs such as ChatGPT, Claude and Gemini are increasingly used to help robots carry out tasks. Researchers warn that secret trigger words could make such robots behave dangerously, and they have developed a defense against the threat. Source: binghamton.edu

llm-d 1.0 brings open, cost-efficient AI inference at production scale

Contributors to the llm-d project have released version 1.0, a major milestone for the open source project. It is designed to help organizations deploy and scale LLM inference in production. Source: newsroom.ibm.com

GoTo releases 27B open-source LLM covering five Indonesian languages

GoTo has released an open-source LLM with 27 billion parameters that covers five Indonesian languages. It expands the Sahabat-AI ecosystem the company launched in 2024. Source: shattered.io

Prompting strategies for LLMs show no stable advantage in benchmark

A comparative benchmark found that LLM prompting strategies were competitive with logistic regression and XGBoost at some operating points. However, they showed no consistently stable advantage over those classical methods. Source: bioengineer.org

Koi: self-calibrating planning for multi-cloud, heterogeneous LLM inference fleets

Deploying an LLM requires choosing the right accelerator, parallelization strategy, engine configuration and replica count. Koi is presented as an evidence-guided, self-calibrating planning approach for managing multi-cloud, heterogeneous LLM inference fleets. Source: alphaxiv.org

LLMjacking: attackers phish cloud billing accounts to run up LLM usage

A single click on a Google Search ad can lead to stolen credentials and a hijacked cloud billing account. Attackers then rack up hundreds of thousands of dollars in LLM usage charges on the victim's account. Source: wiz.io

Inception introduces Mercury 3 as its flagship diffusion LLM

On October 19, 2026, Inception presented Mercury 3, which it calls the most capable diffusion-based large language model released so far. Source: unite.ai

MiniMax unveils M3.1 Flash, claims 70% lower AI agent costs

Chinese AI developer MiniMax Group has released M3.1 Flash, a faster and cheaper update to its M3 large language model. The company says the new model can cut the cost of running AI agents by 70%. Source: siliconangle.com

Inception's Mercury 3 diffusion LLM hits 1,068 tokens per second

Inception has launched Mercury 3, the latest version of its diffusion-based language model. Unlike conventional LLMs, it generates answers through a diffusion process, reaching a speed of 1,068 tokens per second. Source: marktechpost.com

Canadian news publishers sue Anthropic over copyright in Ontario court

Postmedia, Torstar, The Globe and Mail and The Canadian Press filed a lawsuit on Oct. 6 claiming Anthropic infringed their copyright by using their content. The case has been brought in an Ontario court. Source: cbc.ca