Thursday, October 8, 2026
25 stories · 250 sources34호
release · Hacker News · 2건Korean coverage · GeekNews

Anthropic releases small model Claude Haiku 5.5

Claude Haiku 5.5

Anthropic launched Claude Haiku 5.5, a small model built for high-volume, cost-sensitive work. It suits summarization, context compression, database queries, classification and coding-assistant agent tasks, with average run cost about 75% lower than Haiku 4.5.

For high-volume pipelines doing summarization, classification or database queries, you can test whether moving from Haiku 4.5 to 5.5 really cuts average run cost by about 75% and rework which model handles which step.

Today's 72–7

  1. release · OpenAI · 2건Korean coverage · GeekNews

    OpenAI rolls out GPT-6 and Intelligent UI in ChatGPT globally

    GPT-6 and Intelligent UI for everyone

    OpenAI is rolling out GPT-6 globally in ChatGPT, delivering faster responses, to more than 1.2 billion weekly users. Its new Intelligent UI combines text with visuals and interactive elements such as graphics, buttons, forms and charts inside the conversation.

    Since buttons, forms and charts now appear as response formats inside chat, developers who assumed text-only output should revisit their chat interface and integration design.

  2. release · Hacker News

    Mistral Large 4 surfaces on Hacker News

    Mistral Large 4

    A post titled "Mistral Large 4" appeared on Hacker News. The title points to a fourth version of Mistral's Large model line.

    If you already use the Mistral Large line, you can queue it as a swap candidate and prepare to compare output quality and per-call cost against your existing prompts and eval set.

  3. release · Hacker News · 2건Korean coverage · GeekNews

    Google releases open multimodal embedding model EmbeddingGemma 2

    EmbeddingGemma 2

    Google released EmbeddingGemma 2, which places text, code, images, video and audio in the same vector space so they can be searched across each other. It is a 740M-parameter model built on Gemma 4 and released under Apache-2.0, and its modular design lets text-only tasks use just 270M parameters.

    Developers can build cross-modal retrieval or RAG pipelines on one Apache-2.0 model instead of per-modality models, and text-only workloads can load just the 270M-parameter portion to cut memory and inference cost.

  4. research · OpenAI

    OpenAI shares frontier model results on open math problems

    Sharing AI progress in mathematics

    OpenAI published new results on open problems in mathematics from an internal frontier model. It also shared Lean proof formalizations and research details on GitHub.

    Because the proofs are formalized in Lean and posted on GitHub, developers can download and machine-check the model's mathematical claims, which informs how they design verification workflows.

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

    Penguin Mail: open-source Rust email client for Linux with AI

    Penguin Mail – open-source Rust email client for Linux with AI

    Penguin Mail is an open-source Rust email client for Linux that includes AI features. It manages email, calendars and contacts from Gmail, Microsoft and IMAP/POP3 accounts in one app, with a unified inbox and per-account views.

    Developers on Linux desktops who want one client for Gmail, Microsoft and IMAP/POP3 accounts now have an open-source Rust option to evaluate.

  6. industry · Hacker News

    Meta and Microsoft move to reduce employee use of Claude AI

    Meta and Microsoft take steps to reduce employee usage of Claude AI

    Meta and Microsoft are taking steps to reduce employee usage of Claude AI, according to a Hacker News-linked report. The specific measures are not detailed in the headline.

    Teams that built internal coding or workflow tooling around Claude should check how easily they can swap models and how exposed they are to a single vendor as large companies curb external model use.

Releases today2

  1. claude-code v2.1.293Claude Haiku 5.5(claude-haiku-5-5)가 추가돼 Anthropic API의 기본 Haiku 모델이 됐다트래커·GitHub
  2. codex rust-v0.161.0GPT-6.1 Sol이 기본 모델이 되고, 터미널에서 /mcp login으로 MCP 서버에 바로 로그인할 수 있다트래커·GitHub

Rising on GitHubstars gained in a day

  1. morluto/reaReverse engineer anything with agents, from app behavior to native binaries.+4,666
  2. mattpocock/skillsSkills for real engineers, taken from the author's .agents directory.+1,406
  3. cathrynlavery/diagram-designDiagram design skill with 42 diagram types for Claude Code and other coding agents, in self-contained HTML + SVG.+828
This week's repositories →

Also18

  1. Utah to let AI examine patients and prescribe medication without human oversight
    Hacker News·10/7 02:01·120점·댓글
  2. Claude Code’s suggested message feature: I think the real customer is the model
    Hacker News·10/7 03:00·104점·댓글
  3. Multimodal open d1 decision models for the edge
    Hugging Face·01:54
  4. One Model Family, Two Gold-Level Results: Fine-Tuning Nemotron for IOI and IMO
    Hugging Face·10/7 21:45
  5. Introducing Playground: Create and play custom games
    Google AI·10/7 21:00
  6. Muse launches on the iPad
    The Verge AI·00:37
  7. AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model
    HF Daily Papers·10/6 22:56·1점
  8. Can a Cloud-Native Harness Make Agents Reliable Beyond the Desktop?
    Latent Space·10/7 23:10
  9. OpenAI “rogue” agent activities found on Wikimedia projects
    Simon Willison·10/7 09:16
  10. Nous Research confirms it hit $1.5B Valuation, launches AI agents for business users
    TechCrunch AI·05:48
  11. Microsoft releases new Nvidia-chip AI PCs with revamped Windows 11
    TechCrunch AI·05:22
  12. Does better work always mean better workers?
    Google Research·05:19
  13. Helping teens learn, plan, and shape the future of AI
    OpenAI·10/7 21:00
  14. Quoting Victoria Kim
    Simon Willison·10/7 08:58
  15. Radisson Hotel Group brings hotel discovery into ChatGPT
    OpenAI·10/7 16:00
  16. How Jump Trading is scaling quant research with ChatGPT
    OpenAI·10/6 21:00
  17. Anti-Patterns in Software Blogging
    Simon Willison·10/7 23:53
  18. Quoting Jake Boggan
    Simon Willison·10/7 13:47