1. Databricks Makes Chinese Open Model GLM 5.2 Its Default Coding Engine

Databricks. Databricks switched the default model behind its coding tools to the open-source GLM 5.2 from Zhipu after internal testing found it matched Anthropic's Opus on coding tasks at substantially lower cost. The decision is a notable enterprise endorsement of a Chinese open-weight model over premium US frontier models, and it signals that open models are now cost-competitive for production coding workloads. Source

2. OpenAI Finds Roughly 30% of a Popular AI Coding Benchmark Is Broken

OpenAI. OpenAI reported that about 30 percent of tasks in a widely used AI coding benchmark are flawed and said it would stop endorsing the test as a result. For practitioners, the finding is a caution against over-trusting headline coding-benchmark scores and reinforces the need to validate model claims against real workloads rather than leaderboard numbers. Source

3. Ollama Raises $65M as Local-Model Tool Nears 9M Users

Ollama. The open-source tool for running large language models locally raised $65 million and reported growth to nearly 9 million users. The round underscores strong developer demand for local and self-hosted model tooling, and it gives one of the most widely used open-source AI projects meaningful runway to expand. Source

4. New York Times Says OpenAI Concealed Evidence in Copyright Trial

The New York Times. The New York Times escalated its copyright suit by alleging OpenAI withheld tools capable of identifying copyrighted content within training data. The dispute is a bellwether for how courts treat training-data provenance, and its outcome could shape disclosure obligations and liability for model builders across the industry. Source

5. Paris Voice-AI Startup Gradium Raises $100M Seed Backed by Nvidia

Gradium. The Paris-based voice AI startup raised a $100 million seed round with backing from Nvidia and plans to expand into the Bay Area. The unusually large seed reflects continued investor appetite for voice and speech AI infrastructure and strengthens Europe's footprint in the category. Source

6. AI Agent Startup Lyzr Let Its Own Agent Run a $100M Fundraise

Lyzr. The enterprise agent startup said it used its own AI agent to lead a $100 million fundraising process. Beyond the funding itself, the story is a concrete example of an agentic system being handed a high-stakes, end-to-end business workflow, which is relevant to teams weighing how much autonomy to grant production agents. Source

7. FDA Clears iHealthScreen's Autonomous Diabetic Retinopathy Screening Software

iHealthScreen. iHealthScreen received FDA 510(k) clearance for iPredict-DR, software that analyzes color retinal fundus images from an iCare camera to detect more-than-mild diabetic retinopathy. The clearance lets primary care and underserved settings screen for the condition without a specialist, expanding where autonomous AI diagnostic imaging can be deployed in regulated clinical care. Source

8. Questions Mount Over How the US Cleared a Frontier Model as Safe

US government. Reporting probed the still-opaque process by which government reviewers judged a new frontier model safe to deploy, tied to an emerging voluntary pre-release review framework. The piece matters for practitioners tracking how US AI governance and pre-deployment safety review could become a routine gate on model launches. Source

9. Character.AI Enters Interactive Microdramas With Its Own Productions

Character.AI. Character.AI moved into short-form microdrama with original productions that let viewers chat with characters and steer alternate storylines. The launch shows consumer AI platforms pushing into interactive entertainment as a retention and monetization play beyond open-ended chatbots. Source

10. Fundamentum Launches $200M Third Fund Targeting AI and Fintech

Fundamentum. The Indian venture firm launched a $200 million third fund focused on AI and fintech, as co-founder Nandan Nilekani stepped back from his general-partner role. The fund signals sustained institutional capital flowing to AI startups outside the US, particularly across India. Source

11. Analysis Says Coming AI IPOs Could Eclipse 25 Years of Tech Exits

TechCrunch. An analysis argued that forthcoming public listings tied to Anthropic, OpenAI, and SpaceX could generate more value than every VC-backed tech exit since 2000 combined. The framing captures how much capital is concentrating in a handful of AI-era companies and the outsized weight these listings would carry for the broader venture ecosystem. Source