Morning Edition · Tuesday, August 4, 2026
Tech
Alibaba Ships Qwen3.8-Max at 2.4 Trillion Parameters and Promises the Weights Next Week
The model is priced at $2 and $6 per million input and output tokens, roughly a fifth of Anthropic's top tier, and Alibaba says a Max-class Qwen will be downloadable for the first time.
Tech
OpenAI Publishes Lean-Checked Proofs for Ten Open Mathematics Problems From an Unreleased Model
The company puts the compute cost at about $2,000 and released machine-checkable certificates. No result has been through refereed review, and OpenAI is judging the novelty of its own work.
Tech
OpenAI Publishes Internal Messages to Rebut Apple's Trade-Secret Suit
Apple's July complaint names Tang Tan, a former design leader, and Chang Liu, an engineer, and alleges recruiting aimed at unreleased hardware. The case reaches OpenAI's consumer device plans.
Tech
Tencent's Hyra Agent Claims Record Results on 29 of 55 Open Mathematics Problems
The discovery agent runs on the openly released Hy3 model, 295 billion parameters with 21 billion active, and publishes its research artifacts to a public repository.
Tech
OpenAI Details the Full-Duplex Stack Behind GPT-Live, Built in Six Months
The system removes the separate voice-activity detector entirely and trains a speech-native model that listens and speaks at the same time.
Tech
An Open-Source Runtime Streams Mixture-of-Experts Weights From SSD to Run an 80-Billion-Parameter Qwen on a Mac
Swiftlet keeps only the dense core in memory and fetches each routed expert with a single disk read, putting a 35-billion-parameter model on an iPhone at roughly one token per second.
Tech
Meta Sells Its Best Model by the Token After Years of Giving Weights Away
Muse Spark 1.1 reaches outside developers through the Meta Model API at $1.25 and $4.25 per million tokens, undercutting the frontier tiers it is aimed at.
Tech
New Papers Push Back on Paying Frontier Prices to Grade Model Output
One study asks whether cheap open-weight models can judge natural-language mathematical proofs reliably, as forecasters put the model evaluation tools market near $1.15 billion in 2025.
Tech
Meta's Segmentation and Vision Models Move Into Assistive Robotics and National Laboratory Science
The University of Pittsburgh applies Segment Anything and DINO to assistive manipulation, while Lawrence Berkeley National Laboratory uses the same models in early Genesis Mission projects.
Tech
Researchers Propose an Executable Benchmark for the Decisions Agents Make Before They Answer
Two arXiv papers target the choices an agent makes before answering (answer directly, decompose, retrieve, run code, delegate, verify or recover), which drive most of its cost.
Tech
Agent Tooling Turns Toward Traces and Verified Skill Claims
Yandex adds execution-trace analysis to its AI Studio, while an arXiv paper proposes registering agent capabilities on a ledger to close the gap between what a third-party skill claims and what it does.
Tech
OpenAI Publishes a Telco Deployment With Revenue Numbers Attached
The operator Circles reports average revenue per user up 22 percent and churn down 9 percent using the OpenAI API and Codex, figures OpenAI published and no outside party has audited.