Tuesday, October 13, 2026
19 stories · 174 sources39호
release · TechCrunch AI · 2건

Meta releases first frontier models, Muse Wonder and Wonder Full

Meta releases its first frontier model, Muse Wonder

Meta launched Muse Wonder and Muse Wonder Full, which it claims can go toe-to-toe with the industry's best, and which are larger and more capable than the lightweight, low-latency Muse Spark, with higher benchmark scores and better results on complex tasks like coding. Both are available to Muse Code subscribers starting today, Wonder is also accessible through the Meta Model API, and Meta plans to eventually power its Muse agent with Wonder.

Because Wonder is available through the Meta Model API, developers can add it to their shortlist and benchmark it against current models on complex tasks like coding.

Today's 72–7

  1. product · Hacker News · 2건Korean coverage · GeekNews

    Codex CLI installs persistent self-updating background daemon by default

    Codex CLI installs a persistent, self-updating background daemon by default

    A user report says Codex CLI 0.157.0 starts a daemon when an interactive session opens, and it keeps running after all windows are closed. The report asks for daemon_auto_start to become opt-in, and says the daemon keeps an installed platform package of about 374MB in a separate directory.

    If Codex CLI is installed on dev machines, CI or shared servers, check for the lingering background process, the roughly 374MB of disk it uses, auto-updated versions, and the daemon_auto_start setting.

  2. release · Simon Willison

    Python 3.15 ships with explicit lazy imports via PEP 810

    Hello Python 3.15, goodbye 3.10, and now I can use 3.11

    Python 3.15 is out, with a What's new guide curated by 3.14 and 3.15 release manager Hugo van Kemenade. Headline features include PEP 810 explicit lazy imports, where `lazy import json` defers loading until the symbol is first used, and a new unpacking syntax in comprehensions.

    Lazy imports can cut the multi-second startup cost that heavy dependencies like sentence_transformers add to every CLI invocation, so plugin-based tools can rework how they import.

  3. industry · The Verge AI

    Microsoft unveils Maia 300, its third-gen in-house AI chip

    Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs

    Microsoft announced Maia 300, its third-generation in-house AI chip, designed to be the "primary AI inference system" for its products and services. Scott Guthrie said Microsoft expects 40 percent more tokens per megawatt than leading GPU-based systems, with production rollout starting in 2027.

    Teams running inference on Azure may get Maia-based capacity as an alternative to Nvidia GPUs from 2027, so cost-per-token and deployment plans will be worth re-comparing then.

  4. policy · Hacker News

    QGIS bans AI-assisted contributions

    QGIS bans AI assisted contributions

    QGIS has banned AI-assisted contributions, according to a Hacker News post. The headline does not spell out how far the ban reaches.

    Developers sending patches or plugins to QGIS risk rejection if AI tools helped write the code, so they should check how their changes were produced before submitting.

  5. industry · Hacker News

    Research Firm Calls Anthropic IPO an 'Unprecedented Test of Investor Gullibility'

    Research Firm Calls Anthropic's IPO 'Unprecedented Test of Investor Gullibity'

    A research firm has described Anthropic's IPO as an "unprecedented test of investor gullibility." The criticism was shared and discussed on Hacker News.

    Developers building on the Anthropic API may find that financial scrutiny around the IPO could eventually touch pricing and supply stability, which is a reason to review single-vendor dependence and model-switching paths.

  6. product · Latent Space

    Anthropic explains how Dynamic Workflows coordinate Claude agents

    How Anthropic’s Dynamic Workflows Coordinate Claude Agents

    At AI Engineer New York, Anthropic described how its Dynamic Workflows coordinate Claude agents. Drawing on lessons from Unix and Linux, the talk covered how agents can tackle bigger projects without overwhelming developers.

    When designing systems that split large tasks across multiple Claude agents, this offers a reference for how much coordination work can be pushed into the workflow instead of falling on developers.

Releases today1

  1. ollama v0.40.4llama.cpp가 업데이트되어 decision 모델과 EmbeddingGemma 2를 지원한다트래커·GitHub

Rising on GitHubstars gained in a day

  1. morluto/reaReverse engineer binaries and apps through your AI agent, with evidence+45,495
  2. storytold/artcraftAn intentional crafting engine for artists, designers, and filmmakers.+4,106
  3. cathrynlavery/diagram-designDiagram design skill with 42 diagram types for Claude Code and other coding agents, in self-contained HTML + SVG.+2,100
This week's repositories →

Also8

  1. Anthropic's Claude AI submitted false murder tip to Philadelphia police
    Ars Technica AI·03:21·댓글
  2. Why aren't AI companies being charged for hacking?
    Hacker News·10/12 23:52·214점·댓글
  3. Just say no to OpenAI and Anthropic
    Hacker News·03:17·141점·댓글
  4. The gap in shared understanding of LLM capability is widening
    Hacker News·10/12 05:20·113점·댓글
  5. Microsoft president says UK is best placed to drive AI regulation
    TechCrunch AI·00:31
  6. Amazon has over 1 million robot workers. What’s next for human employees?
    Ars Technica AI·10/12 21:43
  7. Russia was not ready for the data center war it started with Ukraine
    Ars Technica AI·02:52·댓글
  8. WindBorne wins a hefty US Air Force contract for its balloon data
    TechCrunch AI·02:19

Korea4

  1. AI 코딩 에이전트는 더 많은 코드를 만들지만, 더 많은 소프트웨어를 만들지는 못함
    GeekNews·05:40
  2. 사카나 AI, LLM 기반 동료 검토 시스템으로 논문 핵심 주장 오류 73% 적발
    AI타임스·02:50
  3. AI 풀스택 ‘준비완료’한 망고부스트 “AI 팩토리로 승부 걸 것”
    THE AI·03:18
  4. "말 한마디로 작동하는 3D 기계 그려"...클로드 코드용 '아나토미' 공개
    AI타임스·03:00