Daily AI Updates: November 17, 2025

November 17, 2025
Overview

SGLang Model Gateway v0.2.3 boosts TTFT and throughput by 20–30% with bucketed routing, adds MinMax M2 function calling and reasoning, streaming tool selection in Chat Completions, tool selection in Responses API, and flexible chat history via PostgreSQL; Grok 4.1 improves conversational intelligence, affective understanding and helpfulness with free access, scores Elo 1722 on creative writing v3 (+600) and cuts hallucinations 3×; DeepAgents refactors on LangChain 1.0 for long‑running multi‑step workflows with middleware; WeatherNext 2 launches 8× faster global forecasting with FGN and 99.9% variable coverage, integrated into Search, Gemini, Pixel Weather, and Google Maps API; Veo 3.1 reaches stable GA on Vertex AI with First/Last Frame and partnerships for minute‑scale cinematic ads; Groq opens an Australia data center with 4.5MW to supply tokens to APAC and accelerate inference.

Main Content

  • SGLang Model Gateway feature updates (v0.2.3): Introduces bucketed routing optimizations that improve TTFT and overall throughput by 20–30%. Adds MinMax M2 function calling and reasoning, streaming tool selection in the Chat Completions API, tool selection in the Responses API, and flexible chat‑history management backed by PostgreSQL and other storages.

  • Grok 4.1 new model release: Frontier model with stronger conversational intelligence, affective understanding, and practical helpfulness—available with free access. Scores Elo 1722 on the creative writing v3 benchmark (+600 over prior gen) and reduces hallucinations by 3×.

  • DeepAgents framework updates: Refactored on LangChain 1.0 to support long‑running, multi‑step workflows (similar to Claude Code). Integrates middleware to handle complex task context and planning.

  • WeatherNext 2 new model release: Global weather‑prediction system with 8× faster generation and coverage for 99.9% of variables (temperature, wind speed, humidity, pressure). Uses a Functional Generative Network (FGN) to produce hundreds of possible forecast scenarios in one step. Integrated into Search, Gemini, Pixel Weather, and the Google Maps API.

  • Veo 3.1 feature updates: Stable GA on Vertex AI with production readiness. Adds First and Last Frame features for stronger narrative control. Partners with Quick Frame and MNTN to generate minute‑scale, cinematic TV and digital ads; adopted by WPP teams for innovation tooling.

  • Groq regional expansion: Australia data center goes live with 4.5MW compute capacity, supplying tokens to APAC developers and accelerating AI model inference.