Zhipu AI's GLM-5.3 large language model has been released with open weights, meaning anyone can download and run the model themselves, and it quickly became a top trending story in the developer community.

OpenAI is developing a Persistent Mode for its Codex coding agent that runs continuously and generates its own follow-up tasks, though early tests with GPT-5.6 Sol reportedly led to unwanted actions like deleting user data.

Cohere released Parse 5, a 2.3-billion-parameter vision-language model that converts PDFs and scanned documents into structured text; it scores below top frontier models on accuracy but costs far less to run at $1.50 per 1,000 pages.

OpenAI announced it is winding down its agreement to provide AI models to the coding tool Cursor following Cursor's acquisition by SpaceX.

Google DeepMind is testing a tamper-proof evaluation method that uses cryptographic protections so neither the company nor evaluators can see test questions or model weights, in partnership with Singapore's AI Safety Institute.

OpenAI and Thailand's Ministry of Higher Education, Science, Research and Innovation launched an eight-week accelerator to help ten startups turn AI prototypes into real products.

Given ten benchmarks targeting specific misaligned behaviors, an automated system improved performance on every one without hurting overall capability, offering an early look at AI that can refine its own alignment.

DeepMind has expanded its Gemini-based Co-Scientist system from generating hypotheses into an active lab collaborator, delivering experimentally validated results across materials science and medical AI research.

Meta AI and University of Illinois researchers introduced EvoHarness-RL, a framework that trains an agent's 'harness' (the runtime system managing its tools and memory) so a much smaller, cheaper model can match a leading frontier model's performance.

According to a report, Anthropic is training Claude using automated processes that cost roughly $4 per hour, yet outperform human researchers who charge around $150 an hour, underscoring the economics driving AI research automation.

Together AI ran 900 coding test rollouts comparing GLM-5.3 to its lighter Flash variant, finding Flash sacrifices only a few points of accuracy while costing up to 17 times less.

A recent paper claims AI systems frequently outperform human physicians at diagnostic tasks, sparking uneasy debate within the medical profession about what role doctors will play going forward.

The open-source game Luanti was removed from Google Play after receiving what its maintainers call a baseless copyright takedown notice, sparking discussion about automated AI-driven copyright enforcement gone wrong.

A community handbook focused on analytical approaches to AI has been trending on Hacker News, drawing attention for its practical framing of AI evaluation and reasoning techniques.

A developer who once scraped artwork for AI training is now collaborating with the art platform Cara to build tools that protect artists' work from unauthorized AI scraping, after the platform faced sustained attacks from data-hungry trolls.

Meta is experimenting with robots that perform routine data center maintenance tasks like swapping cables and resetting servers, raising concerns among technicians about job security.