Large Language Model News

Anthropic's Claude Sonnet 5 model consumes about 40% more tokens per task than its predecessor, nearly doubling real costs despite identical list prices, continuing the company's pattern of hidden price increases. The mo…

A developer benchmarked four Chinese LLM families—DeepSeek, Qwen, Kimi, and GLM—over 30 days using 1,247 prompts across code generation, reasoning, Chinese language, English chat, and vision tasks. DeepSeek V4 Flash emer…

Generative AI has enabled a new wave of sophisticated publishing scams targeting authors. Scammers use large language models to scrape author data from Amazon, Goodreads, and personal websites, crafting personalized emai…

The Trump administration blocked Anthropic's advanced AI models, Fable and Mythos, over national security fears, but the ban has now been lifted. Anthropic has implemented enhanced safeguards and is releasing Mythos 5 to…

Engineer Tejas Chopra founded Headroom Labs after a $287 debugging session revealed that most data sent to large language models is unnecessary. His open-source tool Headroom optimizes token usage by compressing context,…

A developer at Nylas built a CI/CD alert triage agent that uses its own email inbox to receive, cluster, and summarize noisy pipeline notifications. The agent, created via Nylas's custom grant API, replies in the same em…

A developer created a zsh function for agterm that enables per-tab Claude Code session persistence across terminal restarts. The solution uses the tab's unique AGTERM_SESSION_ID as the conversation identifier, allowing e…

Anthropic published an article on recursive self-improvement, exploring how AI systems may increasingly contribute to building better versions of themselves. A developer argues that as AI writes, tests, and reviews code,…

A developer built DevMentor, an AI-powered platform with a Voice-to-Pull Request pipeline that automates the entire software engineering lifecycle. The system converts voice commands into structured tasks, generates code…

A new video tutorial demonstrates how to save a Claude AI chat as a PDF, providing step-by-step instructions for users. The guide aims to help users preserve their conversations with the AI assistant for offline referenc…

Large language model agents are being deployed in process plants to assist operators with fault recovery, proposing actions that are validated externally to ensure safety. The framework aims to reduce economic losses and…

Large language models are evolving from text generators into integrated components in diverse systems, presenting new security challenges across their entire lifecycle. Experts warn that vulnerabilities span eight critic…

OpenAI's GPT-5.6 Pro tier may launch in three distinct variants, according to a benchmark paper, marking a shift from the company's single top-tier model strategy. The move aims to cater to specialized needs and stay com…

GPT-3 is not AGI but language modeling differs from other AI domains because natural language encodes a world model. Scaling language models may drive loss toward Shannon entropy, yielding a usable world model that can b…

A frontend developer at Sima-Land built and published a React date picker npm package called daterly, written mostly with Claude Code and Claude Design. The package prioritizes first-class support for react-hook-form and…

Researchers introduced MIRTH, an enhanced vision-language-action (VLA) framework for robotic control that incorporates dual-scale temporal memory hubs, latent reasoning tokens, and parallel action decoding. The framework…

The Department of Commerce lifted export controls on Anthropic's Claude Fable 5 and Mythos 5 models on June 30, ending an 18-day standoff triggered by a jailbreak discovered by Amazon researchers. The ban had forced Anth…

A new analysis by The Economist finds that the values embedded in leading AI models diverge significantly from those of the general population, raising concerns about the alignment of AI systems with human ethics.

An OpenAI paper reveals three GPT-5.6 Pro models—Luna Pro, Terra Pro, and Sol Pro—breaking with the single top-tier strategy. The Pro tier may now offer speed, throughput, or maximum reasoning power, though it's unclear …

A developer proposes a new AI architectural concept called the 'Semi-Architected Human' model, which shifts from static retrieval to dynamic cognitive autonomy. The model advocates for dialogue-based learning, active res…