AI News 2026-07-31
The Last 24 Days · 12 of 546 stories

Anthropic and OpenAI drop next-gen models while Google advances humanoid robotics and open-weights face pressure.

47 sources swept 546 distinct stories ranked across 24 days 12 worth your time
  1. Introducing Claude Opus 5

    Anthropic's new flagship model sets a new state-of-the-art benchmark for frontier intelligence and complex reasoning tasks.

    Anthropic··Models & releases·lab·4 outlets
  2. Advancing the price-performance frontier with GPT-5.6

    OpenAI's GPT-5.6 release significantly lowers API costs, making high-tier reasoning economically viable for production-scale enterprise workflows.

    OpenAI··Models & releases·lab·3 outlets·openai.com
    Also covered bySimon Willison
  3. Gemini Robotics 2 brings whole body intelligence to robots

    Gemini Robotics 2 shifts robot control from isolated manipulation to unified, whole-body humanoid coordination using a single model.

    Google DeepMind··Models & releases·lab·2 outlets·deepmind.google
  4. Kimi K3: The open-weights escalation

    Kimi K3's release intensifies the open-weights race, challenging Western dominance in high-context frontier models.

    Interconnects··Business & policy·newsletter·2 outlets·interconnects.ai
    Also covered byTLDR AI
  5. Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber

    Google's new Flash models optimize speed and cost efficiency for high-throughput, low-latency production applications.

    Google DeepMind··Models & releases·lab·2 outlets·deepmind.google
  6. Discovering cryptographic weaknesses with Claude

    Anthropic demonstrates LLMs identifying real-world cryptographic vulnerabilities, signaling a shift in automated vulnerability research.

    Anthropic··Safety & security·lab·2 outlets
    Also covered bySimon Willison
  7. OpenAI and Hugging Face partner to address security incident during model evaluation

    The joint security response highlights critical vulnerabilities in model evaluation sandboxes and establishes new defense protocols.

    OpenAI··Safety & security·lab·2 outlets·openai.com
  8. Judge approves $1.5B Anthropic settlement for pirated books used to train Claude

    A massive $1.5B copyright settlement establishes a costly legal precedent for training frontier models on copyrighted data.

    Hacker News AI··Business & policy·HN·568 pts·apnews.com
  9. Welcome Inkling by Thinking Machines

    Thinking Machines' Inkling model introduces novel reasoning architectures to the open-source ecosystem via Hugging Face.

    Hugging Face··Models & releases·lab·4 outlets·huggingface.co
  10. LFM2.5-Encoders for Fast Long-Context Inference on CPU

    LFM2.5-Encoders enable fast, long-context document processing on standard CPU hardware without requiring expensive GPU clusters.

    Hugging Face··Infrastructure & chips·lab·2 outlets·huggingface.co
    Also covered byMarktechpost
  11. Native-speed vLLM transformers modeling backend

    Integrating native-speed transformers directly into vLLM significantly reduces inference latency and improves serving throughput.

    Hugging Face··Infrastructure & chips·lab·huggingface.co
  12. 6 months to live for open models

    An analysis of the economic viability of open-source AI highlights the unsustainable cost gap facing non-commercial model creators.

    Interconnects··Safety & security·newsletter·interconnects.ai