AI coding models

A recorded talk by Greg Kroah-Hartman, the longtime Linux kernel developer and maintainer of its stable branch, examines how large language models are changing software security. The discussion covers the risks and practical questions of using AI-generated code in critical infrastructure. It is drawing attention from developers weighing how AI tools affect the integrity of open-source projects.

A new analysis from Manifold Security examines how large language models answer questions about whether code is malicious, finding they frequently invoke moral framing rather than purely technical judgment. The post is drawing attention on Hacker News, where commenters are debating whether moralized reasoning makes AI security assessments less reliable or more explainable, and what it means for using models in malware triage.

A new essay argues that Claude Code's suggested-message feature is best understood as serving the AI model rather than the human user. The author, Zohaib, contends the auto-generated prompts help the model steer conversations and guide coding sessions more effectively. Readers on Hacker News are debating whether such suggestions primarily benefit developers' productivity or the model's own performance.

A developer behind the Wagtail content management project has published a write-up on using GLM 5.3 Flash as a coding assistant for a full month. The post describes hands-on experience with the model in day-to-day programming work and is drawing attention among developers weighing newer, faster AI models for their workflows.

A music video titled "LGTM (Looks Good to Me)" has been created using Anthropic's Claude Opus AI model, and it is drawing attention on Hacker News. Commenters are amused by the novelty of an AI-generated song built around the familiar code-review phrase "looks good to me," treating it as a lighthearted showcase of current AI music and video generation tools.

Eric S. Raymond, co-founder of the open-source movement, argues that closed-source software is effectively finished. He says AI-assisted reverse engineering has become so capable that proprietary code can no longer be meaningfully kept secret, undermining the business model behind closed-source products. The claim is drawing attention and debate in developer and tech communities.

A new discussion in developer circles claims that AI coding agents perform better when they do not write their own tests, contradicting the common assumption that self-testing improves code quality. Developers are debating why test generation may mislead agents, with some saying letting models grade their own work invites blind spots rather than catching bugs.

AI company Cognition has introduced SWE-2, a new software engineering model it says pushes the Pareto frontier, meaning it improves the trade-off between capability and cost or speed. The announcement positions the model as a step forward for automated coding and engineering tasks, and it is drawing attention from developers and AI observers tracking rapid progress in coding agents.

Anthropic has released Claude Haiku 5.5, its most affordable model to date, extending its Claude lineup with a lower-cost option aimed at high-volume, cost-sensitive AI applications. The announcement is drawing attention across the tech sector, with discussion focused on pricing, speed, and how the budget model compares with competitors' small models for everyday coding, summarization, and automation tasks.

A new essay examines how Clojure fits into software development shaped by large language models. The author, known in the Clojure community, discusses whether the language's simplicity, functional design and stable syntax make it well or poorly suited to AI-assisted coding. Readers are sharing and debating the argument in programming circles.

A new essay examines how Clojure, the Lisp dialect on the JVM, holds up as large language models reshape software development. The piece weighs Clojure's simplicity, stable syntax and functional design against the way AI coding assistants are trained mostly on more mainstream languages, sparking debate among developers about the language's future relevance.

Chinese AI firm StepFun has released Step 5 Preview, which appeared as a fresh listing on a new-models tracking channel shortly after publication. The model is being tagged in open-source and developer communities alongside topics like AI, coding and web development. People are watching to see whether Step 5 advances StepFun's position among open-weight model makers.

Google is back in the spotlight after announcing its Gemini 4 flagship AI model following months of delays, and searches for 'antigravity' are climbing in the United States. The term is associated with Google's AI developer tools and coding agents, which have drawn heavy interest with each Gemini release as the company pushes its agent-based approach to software development.

Anthropic has released Claude Haiku 5.5, described as its cheapest model while offering enhanced coding and computer-use capabilities. The release is drawing attention across multiple countries, with coverage highlighting the combination of low cost and improved technical performance as a competitive move in the AI market, particularly against rivals offering budget-tier coding assistants.

A software development kit that applies AI-assisted coding to wireless microcontrollers has moved into beta testing. The release adds an open-source community for developers and connects to enterprise MLOps tooling, aiming to simplify how embedded engineers build and deploy machine learning models on low-power wireless chips.

JetBrains has released Mellum 2.1, a 12-billion-parameter mixture-of-experts open model designed for coding agents. Reported by MarkTechPost, the release positions the model as a lightweight, open option for developers building AI-assisted programming tools, continuing JetBrains' move into open AI models for software development.

Open-source pioneer Eric S. Raymond argues that closed-source software is effectively finished, because AI tools make reverse engineering proprietary code cheap and easy. Writing about AI's impact on the software industry, he says vendors can no longer keep their programs secret in any meaningful way, and that the economics that sustained proprietary software are collapsing as machine-assisted analysis spreads.

A benchmark called BankSafe-Bench has been submitted to the Kaggle Benchmarking Challenge. It measures how well AI coding models handle realistic financial software engineering tasks, focusing on whether models can safely write code for banking-related applications. The submission is being shared among developers and AI enthusiasts interested in evaluating the reliability of code-generating systems in high-stakes financial settings.

Anthropic's Claude Haiku 5.5 model is now usable in Claude Code, the company's command-line coding assistant, but only from version 2.1.293 onward. Developers are being told to run the claude update command to get access. The model can be used directly or assigned as a subagent, letting it handle smaller tasks within larger agent workflows.

A question on Hacker News is asking which frontier AI models are capable of performing code security reviews. The discussion seeks recommendations on which large language models can reliably audit code for vulnerabilities. With few replies so far, the thread reflects a broader interest among developers in using AI tools for security analysis.

Anthropic's Claude Haiku 5.5 is being presented as its most affordable model, with claimed improvements in coding and computer-use capabilities. Coverage so far focuses on the price-to-performance pitch and what the upgraded computer use means for automating tasks. Reaction is still limited, with details on benchmarks, pricing tiers and availability not yet fully reported.