A November 10, 2026 post outlines decision models: models that assign probabilities to predefined answers instead of generating free text. The post explains choice, score and noul primitives and shows examples, API support and tooling updates.
Researchers behind XYBench release an 8,115-question benchmark that tests whether assistants detect users’ mistaken assumptions and prioritize underlying goals. Results show several large models more often address the stated request than the user’s actual intent.
LangChain has closed a $300 million Series C at a $3 billion valuation led by CapitalG, the company announced. LangChain says it has shifted from open source tooling to agent management and reports 7,000 LangSmith customers, including Nvidia and Toyota.
A walkthrough published November 13, 2026 describes a beverage can pick-and-place task built in Isaac Sim 6.2 using the UR17G arm, Robotiq 2F-140 gripper and assets from SimReady Central on Hugging Face. The authors report profile validation and agent-driven controls (GPT-6 Sol example).
Researchers say exported Blackwell GPUs appear in weights of leading Chinese open models, and allegations link Megaspeed executive Alice Huang to chip smuggling. Epoch AI estimates about two million advanced chips entered China in 2024–2025; investigations and export-control scrutiny are ongoing.
The llama.cpp and ggml projects published v1.0.0 releases that formalize API and ABI boundaries and commit to semantic versioning. The announcement lists distribution channels, hardware partners, and continued support for 150+ model architectures.
Meta has opened a Muse feature that lets the AI place calls to U.S. businesses, now available to paid subscribers after testing began in September. Subscription plans start at $20/month (500M tokens/week) and $100/month (3B tokens); similar options exist from Instinct and Google’s Gemini.
Tether AI Research released BotPsy, a 0.8B-parameter model based on Qwen3.5-0.8B for on-device function calling and agentic tasks. The team reports it outperforms models up to 1.2B on most tool-use benchmarks and will be published under Apache 2.0.
Google introduced the Gemini Decisions API and the Gemini Decision 1 model to score up to 1,000 decision questions in a single multimodal pass. Google reports 150–250 ms end-to-end latency and a price of $0.10 per 1M input tokens, with no output-token charges.
Hugging Face announced on November 10, 2026 that organizations can link their Hugging Face organization to a Google Cloud account so Hub usage appears on the Google Cloud bill. The integration is available via the Google Cloud Marketplace and covers subscriptions, compute and storage.