AI News: Coding

Running the World's Best AI Models on Your Own Computer

A new tool makes it easier to run the most advanced language models on your own device, without relying on the cloud. This could change the way people use AI in everyday life.

Anthropic Launches Claude Science for Scientific Research Automation

Anthropic, a leading AI company, has unveiled Claude Science, a cutting-edge product designed to revolutionize scientific research. This new tool can autonomously perform complex tasks with high-level instructions, freeing up researchers to focus on groundbreaking discoveries.

New AI Benchmark Helps Companies Migrate to Java Framework

A new AI benchmark called ScarfBench is designed to help companies migrate their software to the Java framework. This tool can speed up the process and reduce errors by evaluating and comparing different AI agents.

AI Model Runs Nonstop for 19 Days, Generates Code from Scratch

AI has been pushed to its limits with a new benchmark test. An AI model named Claude Opus 4.7 ran for 19 days straight, generating code from scratch without seeing the original code. This achievement comes with a hefty price tag.

Claude Code Exposes Developers to Hidden Malware on GitHub

** A security flaw in the AI coding tool Claude Code allows hackers to secretly install malware on a developer's machine by hiding it in a GitHub repository. This makes it hard for scanners to detect.

OpenAI Tackles Open Source Bugs with AI-Powered Initiative

OpenAI is launching a new initiative to help find and fix bugs in open source code. This AI-powered project could make software safer and more reliable.

Engineers Now Outpacing Product Managers Thanks to AI

Engineers are using AI to become three times more productive, but this shift is creating a new bottleneck: deciding what to build. Companies are struggling to keep up with this change, and product management is becoming the new bottleneck.

Alibaba's New AI Model Breaks Agent Performance Records Across Seven Benchmarks

Alibaba's researchers have made a breakthrough in AI by training a model that predicts what environments return, not what agents should do. This new approach has improved agent performance across seven benchmarks.

Xiaomi's HarnessX rewrites its own AI scaffolding mid-task — and smaller models gain the most

AWS Tackles Flaws in AI Agents with Two New Services

Amazon Web Services (AWS) has launched two new services to address the issues of AI agents lacking business context and security. These AI agents can write code quickly, but often make mistakes. The new services aim to fix this problem.

AI App Builder Creates Functional Program in Just 5 Minutes

A new AI app builder claims to create functional programs in just 5 minutes. This is a significant improvement over traditional coding methods.

New AI Framework Boosts Performance by 2.5 Times with Less Computation

Researchers have developed a new AI framework called Arbor that significantly improves the performance of AI-driven research and optimization. This breakthrough is expected to revolutionize the way companies use AI to automate complex engineering tasks.

Xiaomi's new open source, agentic AI coding harness MiMo Code beats Claude Code at ultra-long, 200+ step tasks

Xiaomi's MiMo team has released an open-source AI coding assistant called MiMo Code, which outperforms Anthropic's Claude Code on long tasks. MiMo Code uses a unique approach to remember and recall context, making it a more reliable coding partner.

Vibe Coding Limits: Why AI-Generated Systems Lose Context Over Time

AI coding agents are rapidly changing data engineering, but they come with a hidden cost. Enterprise data platforms struggle to maintain consistency and visibility as context and logic get scattered across prompts and generated code.

Open-Source Coding Agent Released for Software Engineering Tasks

A new open-source coding agent is now available for software engineering tasks. This agent, called North Mini Code, can perform complex coding tasks like code review and architecture mapping on its own.

NanoClaw and JFrog launch 'immune system' to block AI agents from downloading malicious code

** Two major tech companies team up to keep AI assistants safe from cyber threats. They've created a new security system to prevent malicious code from being installed on autonomous agents.

Open model Kimi K2.7 Code undercuts GPT-5.5 and Claude by up to 12x on price per token

Moonshot AI's new open-weights model, Kimi K2.7 Code, outperforms GPT-5.5 and Claude in price, not necessarily quality. This affordable model is built for programming and offers an alternative to more expensive options.

New AI Model Cuts Thinking Time But Practitioners Question Benchmarks

Moonshot AI releases Kimi K2.7-Code, a new AI model that reduces thinking tokens by 30%. However, experts question the model's performance on independent benchmarks and whether it truly outperforms other models.

Google's AI Research Tool Gets Major Upgrade

Google is upgrading its NotebookLM research tool, giving it its own cloud computer and advanced features. The tool can now run code, search the internet, and even beat its previous version in internal tests.

Apple Finally Catches Up with AI Promises

Apple's annual developer conference kicked off with bold AI promises, but it seems they were more about catching up than pushing the limits. The company unveiled a new "Siri AI" that aims to improve its virtual assistant capabilities.