Skip to content
AI.info

The Pulse · In brief

In brief · Wednesday 18 November

The day’s news in brief, Wednesday 18 November: 6 briefs, each written from its original page and checked against it.

6 briefs from 6 sources; the last filed at 02:09 UTC.

New model 1 · Product 2 · Research 1 · Partnership 1 · Compute 1

Written with AI · fact-checked

In brief · Wednesday 18 November

  1. Research

    FastFrames authors report GPU timings for two model sizes

    The paper’s authors say the 1.3B version creates a frame in 36 ms with one B200 GPU, whereas the 14B version takes 45 ms per frame on an 8x B200 GPU node. They report that the smaller model reaches the quality of a state-of-the-art system running nearly 8x slower, while the 14B model delivers significantly better visual quality and realism.

    Source · arXiv · · #

  2. Product

    Guidepoint brings expert research workflows into major AI platforms

    Guidepoint says its new MCP capability connects Guidepoint360 with Claude, ChatGPT, Gemini, and Perplexity, letting research teams conduct primary research and manage expert consultations in their chosen AI environment. The company says researchers can query its Library of 150,000+ transcripts, with 5,000+ added every month, and trace answers to source transcripts.

    Source · Webdisclosure · · #

  3. Partnership

    Astera Labs names Hyve among its first Orion rack partners

    Astera Labs says Orion pairs pre-tested CPU and accelerator configurations, and that Hyve Solutions is among the first companies developing racks based on the platform. Astera Labs says OEM and ODM partners are expected to offer Orion reference and system designs starting in Q1 2027.

    Source · Astera Labs · · #

  4. Compute

    Dell and AMD add server and rack-scale options to AI Factory

    Dell says it is extending Dell AI Factory at SC26 with the PowerEdge M8825, an AMD EPYC-based server. The company says modular options include PowerEdge XE7745 and R7725 servers with AMD Instinct MI350P PCIe GPUs, validated for AMD Enterprise AI Suite, Red Hat OpenShift Container Platform and Red Hat AI Enterprise. Dell says its AMD Helios Rack-scale Solution can deliver up to 2.9 exaflops using 72 AMD Instinct MI455X GPUs.

    Source · Dell Technologies · · #

  5. New model

    Microsoft introduces its MAI Realtime 1 speech-to-speech model

    Microsoft says MAI Realtime 1, its first model that converts speech directly into speech, is designed for enterprise conversations and costs from $0.01 per audio minute. Microsoft says it performs better than GPT Realtime 2.1 on quality, speed and price in public benchmarks and Microsoft product evaluations, and sits on the performance-price Pareto frontier in Big Bench Audio speech reasoning and the Tau-Voice agentic task completion benchmark.

    Source · Microsoft AI · · #

  6. Product

    SiMa.ai releases Palette 3.0 for Physical AI workloads

    SiMa.ai announced Palette 3.0 on November 17, 2026, describing it as a software update intended to improve performance across Physical AI models. The company says the release lets multiple models run at once for machine reasoning and action, and raises model tokens per second by up to 3.5x while staying within a low-power budget.

    Source · SiMa.ai · · #

Articles of the day

How briefs are made: each is written from one primary page our reader opened, checked sentence by sentence against that page by a second model, and read before it is published. How we write

Every brief is machine-written and fact-checked, and says so (Regulation (EU) 2024/1689, art. 50(4)).