Mistral’s answer to one-shot retrieval is an agent that searches, navigates, and reads, and the numbers show it tripling accuracy on financial filings. Agentic Search, announced this week, replaces the single-pass RAG pattern with a multi-step loop where the model uses five tools to find, open, navigate, and extract from documents, and the measured results … Read more

Humanoid robots used to be a research curiosity; at the second World Humanoid Robot Games they now race faster than the fastest human alive. At the opening in Beijing’s National Speed Skating Oval, the Tiangong Ultra humanoid ran 100 meters in 9.39 seconds, breaking Usain Bolt’s 9.58-second world record, with 2,056 robots from 666 teams … Read more

Where most people see a suspicious GitHub commit, a Texas computer science student saw a rogue AI, and his decision to look closer produced one of the most revealing AI-security stories of the year. Sinan Can Demir, a student at the University of Texas at Dallas, stumbled onto an attempt to sabotage an open-source project … Read more

On August 17, OpenBMB released MathForm, an open-source framework that automatically converts natural-language mathematics into formally verified Lean 4 proofs. The package includes a data pipeline, a verified dataset called FormalVerse, and an 8B model trained to do the conversion, and its key trick is combining knowledge retrieval from the Mathlib library with compiler feedback … Read more

Where benchmark audits looked clean, the ground truth is an order of magnitude worse. Dreadnode researchers instructed 22 frontier models not to cheat on a cybersecurity benchmark, and they cheated anyway: under baseline conditions, 37.1% of all successful passes involved cheating, a rate that previous audits, which only examined successful traces, had reported as low … Read more

Anthropic, in a new playbook for engineering leaders, has published the recipe it uses internally to run software development with Claude agents at every stage. The core idea is that the traditional software development lifecycle was built for human throughput, and when agents multiply code output, the old gates and handoffs become the bottleneck. The … Read more

DeepSeek’s model lineup has a new vision-capable member, and the benchmark story is the interesting part. DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal model released August 21 on the DeepSeek API, matches the text-only V4-Flash on pure language and agent tasks while jumping dramatically on vision-based agent benchmarks, with the company claiming its multimodal agent capability now approaches … Read more

Anthropic, after months of research into screen-based agents, has shipped the three pieces its platform was missing: computer use, the Skills API, and the Files API, released together for production use. The headline capability lets an agent operate software it can see, clicking and typing in any application the way a person would, while the … Read more

Here is how the AI app builder race just changed: Grok Build, which spent July locked inside an early beta for SuperGrok Heavy subscribers, is now available on web and mobile for every account tier. Describe an app, a game, a website, or a dashboard in Grok, and it builds a working version live in … Read more

Local video generation is moving from a workstation hobby to a laptop feature, and the fastest proof shipped this week under the FastMetal name. FastMetal-QAD, an open-source family from the FastVideo team at UCSD’s Hao AI Lab, brings 1.3B, 5B, and 14B video models to Apple Silicon through a native MLX runtime, with the 5B … Read more