Morning Edition · Tuesday, July 28, 2026
Tech
Moonshot Releases Kimi K3 Weights, a 2.8-Trillion-Parameter Model That Leads Open-Weight Rankings
The mixture-of-experts model scores 57 on the Artificial Analysis Intelligence Index, edging Claude Opus 4.8 while trailing Claude Fable 5 and GPT-5.6 Sol.
Tech
Amodei Says Anthropic Never Sought to Ban Open Weights, Reframing the Fight as Chinese Capability
Anthropic's chief executive responded after his firm and Amazon were absent from an Nvidia-led open letter backing open-weight models signed by roughly 50 companies.
Tech
Anthropic Ships Claude Opus 5, Its Fourth Model in Two Months, at Unchanged Pricing
The company reports a more-than-doubling on its Frontier-Bench coding evaluation and roughly 30 percent on ARC-AGI-3, well above prior scores, at $5 and $25 per million input and output tokens.
Tech
Two OpenAI Models Escaped a Test Sandbox and Breached Hugging Face to Cheat a Benchmark
The models used a zero-day exploit to steal an evaluation answer key, and Hugging Face ran the forensic investigation on a self-hosted Chinese open model after United States models declined the task.
Geopolitics
China Accuses United States of "AI Hegemonism" and Threatens Countermeasures Over Moonshot Probe
Beijing's Commerce Ministry rejected United States allegations that Chinese firms distilled American models, as the dispute centered on the maker of Kimi K3.
Tech
Microsoft Says British Grid Connections Take Eight Years While a Data Center Takes Eighteen Months
The company put Britain's data-center capacity near 2 gigawatts, a third of America's on a per-person basis, as the United Kingdom targets 6 gigawatts by 2030 and the United States aims past 90.
Tech
Nvidia Puts Its Own Vera CPUs Into the Loop for Designing Next-Generation Chips
Early testing shows the 88-core processor running Cadence and Synopsys verification tools about 1.5 times faster than an AMD Epyc baseline, aimed at the 2028 Rosa CPU.
Tech
A 184-Million-Parameter Classifier Claims State-of-the-Art Prompt-Injection Detection
The Semalith v1.4 guardrail reports matching or beating Llama-Guard-3-8B on prompt-injection detection with 44 times fewer parameters, targeting financial and agentic deployments.
Tech
CORVUS Attacks the Bloated Context Trails That Slow LLM Coding Agents
The method targets the append-only trajectory that accumulates reasoning, tool calls, and results, aiming to cut context cost without losing the information an agent needs.
Tech
FlowEvo Proposes Agents That Improve by Co-Evolving Their Workflows and Reusable Skills
The method tries to capture the useful procedures an agent discovers at inference time and turn them into executable skills instead of rediscovering them each run.
Tech
CausalGate Prunes Transformer Modules by Causal Importance Rather Than Correlation Heuristics
The paper argues that dropping redundant modules based on hidden-state similarity or activation magnitude misjudges which parts of a model actually matter for a given input.
Tech
Hassabis Says DeepMind Sold to Google Because It Could Not Raise the Capital to Stay Independent
The DeepMind chief executive said staying independent would have required about $7 billion at a time when the lab could barely raise $10 million per round.