AI News 2026-08-03
New To You

Alibaba releases 2.4‑trillion‑parameter Qwen3.8‑Max, rivaling top US models

50 sources swept 451 distinct stories from the last 1 days 88 already covered, held back 12 worth your time

  1. Alibaba’s open-weight Qwen3.8-Max takes on long-horizon AI tasks with 2.4 trillion parameters

    Qwen3.8‑Max can autonomously build software, reproduce research, and run a simulated e‑commerce business, matching Western models in internal benchmarks.

    The Decoder · · Models & releases · press
  2. China’s Alibaba takes another swipe at America’s AI supremacy

    Alibaba claims Qwen3.8‑Max is its most capable model yet, positioning it against Anthropic, OpenAI and domestic rivals like Moonshot AI.

    The Verge AI · · Models & releases · press
  3. OpenAI's Unreleased Model Astra Solves Ten Major Open Mathematics Problems

    OpenAI’s unreleased Astra model solved ten major open mathematics problems, demonstrating genuine mathematical reasoning beyond earlier LLMs.

    Don't Worry About the Vase · · Models & releases · newsletter · thezvi.substack.com
  4. GEM Training: How Meta Doubled the Efficiency of Its LLM-Scale Ads Foundation Model

    Meta’s Generative Ads Recommendation Model (GEM) doubled training efficiency to 20‑25% MFU while scaling FLOPs 4× in a year.

    Meta Engineering (ML) · · Infrastructure & chips · lab · engineering.fb.com
  5. How we built a realtime system for responsive voice AI in six months

    GPT‑Live introduces a turn‑less speech architecture that reduces latency, enabling continuous, natural voice interactions.

    OpenAI · · Models & releases · lab
  6. From weeks to minutes: How Formula 1® uses agentic AI on AWS to accelerate data operations

    Formula 1 used AWS Bedrock AgentCore to cut data source onboarding from weeks to about 40 minutes, accelerating MarTech decisions.

    AWS Machine Learning · · Agents & tooling · lab · aws.amazon.com
  7. Automated Reasoning policy refinement in Amazon Bedrock

    Amazon Bedrock now automatically refines Automated Reasoning policies by diagnosing failing tests and proposing formal‑logic fixes.

    AWS Machine Learning · · Safety & security · lab · aws.amazon.com
  8. Europe’s AI labeling and transparency rules are now in effect

    EU AI Act transparency obligations took effect, requiring disclosures for chatbot and deep‑fake interactions under threat of fines.

    The Verge AI · · Safety & security · press
  9. How to Run Isolated Tenant Kubernetes Clusters on Shared GPU Infrastructure

    NVIDIA outlines a pattern for isolated tenant Kubernetes clusters on shared GPU infrastructure, reducing cross‑team interference.

    NVIDIA Developer · · Infrastructure & chips · lab
  10. Orchard: An open framework for scalable agentic AI

    Orchard provides an open‑source framework that lets researchers reuse environments and data pipelines to train agentic AI across domains.

    Microsoft Research · · Agents & tooling · lab
  11. China's MiniMax H3 is the first open model to top an AI video ranking

    MiniMax’s 33‑billion‑parameter H3 model topped AI video rankings in multiple categories, generating multi‑second clips with stereo sound.

    The Decoder · · Models & releases · press
  12. Two teams solved the same quantum crypto problem using GPT-5.6 just three hours apart

    Two independent teams used OpenAI’s GPT‑5.6 to solve the same open quantum cryptography problem within three hours.

    The Decoder · · Research · press